thesis_underwater swarm sensor networks
TRANSCRIPT
Underwater Swarm Sensor Networks: Applications, Deployment, and
Medium Access Communication Protocols
By
Gunilla Elizabeth Burrowes BE, MPhil
Doctor of Philosophy
January 2014
ii
Statement of Originality The thesis contains no material which has been accepted for the award of any other degree or
diploma in any university or other tertiary institution and, to the best of my knowledge and belief,
contains no material previously published or written by another person, except where due
reference has been made in the text. I give consent to the final version of my thesis being made
available worldwide when deposited in the University’s Digital Repository, subject to the
provisions of the Copyright Act 1968.
……………………………………… Gunilla Elizabeth Burrowes
iii
Dedication
This thesis is dedicated to my parents
who I am indebted to for giving me their love of learning and life
To the memory of my father,
Richard (Dick) Ranson BE Hons(Syd)
who inspired and supported me to become an engineer
and
To my mother,
Kerstin Ranson
who continues to inspire me everyday with her determination and affection.
iv
Acknowledgements
A piece of work such as this thesis is never done alone and I have so many people to thank and
acknowledge.
Firstly, to my supervisors who has supported me to achieve this goal. It has taken so much
longer than planned for many reason and I would like to particular thank my supervisor A/Prof
Jamil Khan for staying with me throughout this journey and always being available when
needed. Thank you also to Dr Jason Brown as my co-supervisor who has provided invaluable
support with OpNet, and was always willing to help with ideas and enthusiasm.
To my family, who if they were not by my side, I could not and maybe would not have completed
this thesis. In particular, to my husband Darren, thank you so much for all your love and support
over the years and the encouragement to keep going. To Edward and Ingrid, my beautiful
children, who are my pride and joy; thank you for your understanding during the many times that
I could not be there for you and for the inspiration that you gave me to succeed.
And also to my parents who have always believed in education. I will be eternally grateful for
their continual interest and faith in what I have done in my life that has lead me to the path of
taking on the challenge of a PhD. Thankyou to my mother, Kerstin Ranson, for her wisdom in
my life and to my brothers Eric and David and sister Caroline and their families who I am so
lucky to share my life with.
I am also indebted to my many colleagues who have had to take on extra work to allow me time
to continue to study. To my business partner, Dr Mark Toner, thank you for continuing the
business almost without me and for your words of encouragement and humor. Thank you also
to Prof David Dowling for all your advice and words of support.
And to all my wonderful friends, thank you for being there to share a coffee and a laugh.
And finally to my study buddy, my wonderful boxer dog, Ronia who passed away last Christmas
before I could finish this work.
v
Abstract Our oceans are vast and remain mostly unexplored. Advances in underwater technology have
enabled exciting new applications for underwater wireless sensing and monitoring of the
environment, fauna, flora, and human activity. The 'game changer', however, for future
developments will be when swarms of mobile vehicles are able to undertake autonomous
missions as they will increase the usefulness and ability to begin extensive sampling of the
earth's oceans to gain an insight into this unknown world.
Current solutions have been built around static sensor networks and single ROV’s (remotely
operated vehicles) and single AUV’s, which have been typically sparsely deployed. The growth
of underwater operations will require data communication between various homogeneous and
heterogeneous underwater networks and surface based equipment.
This thesis has focused on the communication requirements and medium access control (MAC)
algorithms for groups of AUV’s operating in close proximity to each other in a swarm-like fashion
as an underwater swarm sensor network (USSN). An investigation into the various applications
that would benefit from using a swarm of AUV’s has lead to the classification of Non-time
Critical Missions, for mapping and surveying for example and Time Critical Missions for using
real-time payload data collection for searching for an object or target. This leads to two topology
configurations, a Bus Topology and Cluster Topology respectively that requires different Quality
of Service boundaries and MAC methods.
The requirement to operate vehicles at very close-range has meant an investigation into the
atypical short-range underwater acoustic channel and the spatial-temporal diversity that
acoustic communications between devices underwater create which is different from long-range
underwater acoustic communications and very different from RF communications in terrestrial
settings. This work has also studied the data exchange needs of swarming algorithms with a
focus on bio-inspired algorithms that can be used in a group of AUV’s to facilitate the formation
of vehicles in particular the Cluster Topology.
To maintain swarm synchronisation in both Topologies real-time communication is required in a
fully connected but distributed group of underwater vehicles (AUV) operating in an USSN. Two
MAC layer protocols were developed for the different application areas: “Adaptive Token Polling
MAC (ATP-MAC)” has used an adaption of a token polling ring to provide a decentralised
distributed MAC protocol for the Non-time Critical Missions and “Adaptive Space Time – Time
Division Multiple Access (AST-TDMA)” protocol that utilising a token to trigger time divisions
between vehicles rather than a clock used in TDMA is a fully distributed decentralised algorithm.
Both protocols are designed to effectively use a single channel broadcast acoustic environment
while incorporating a method to handle the spatial-temporal characteristics experienced
underwater. They allow operations to be independent of time synchronization between vehicles
and require no prior knowledge of propagation delays.
vi
Analytical results presented in this thesis show both the AST-TDMA and ATP-MAC protocols
exhibit substantial advantages over the conventional TDMA protocol for the applications that
they are designed for. It is shown that the new adaptive protocols outperform TDMA in their
ability to disseminate time-sensitive information in a timely manner and therefore allow much
higher densities of vehicles to operate in swarm-like networks in both the Bus and Cluster
Topologies studied.
The AST-TDMA protocol operations in a non-ideal underwater communication channel have
also been simulated and the results are presented and analysed. This non-ideal channel
includes the simulation of noise and reverberation models. A proposed new type of
reverberation, Swarm Reverberation, has also been introduced and incorporated in the
analysis.
vii
Associated Publications
The following publications are associated with work in this thesis:
[1] Burrowes, G.E., Khan, J.Y., “Adaptive Token Polling MAC Protocol for Wireless
Underwater Networks.” International Symposium on Wireless & Pervasive Computing.
Melbourne, 2009.
[2] Burrowes, G.E., Khan, J.Y., “Investigation of a Short Range Underwater Acoustic
Communication Channel for MAC Protocol design”, 4th International Conference on
Signal Processing and Communication Systems (ICSPS) 2010, IEEE Conference
Publications, Digital Object Identifier: 10.1109/ICSPCS.2010.5709665
[3] Burrowes, G.E., Khan, J.Y., “Short-range Underwater Acoustic Communication
Networks.” In Autonomous Underwater Vehicles, by Nuno A Cruz, 173-198. Croatia:
InTech, 2011.
[4] Burrowes, G.E., Brown J., Khan, J.Y., "Adaptive Space Time - Time Division Multiple
Access Protocol (AST - TDMA) for an Underwater Swarm of AUV's". IEEE OCEANS,
Bergen, June 2013
[5] Burrowes, G.E., Brown J., Khan, J.Y., "Impact of reverberation levels on short-range
acoustic communication in an Underwater Swarm Sensor Network (USSN) and
application to transmitter power control". IEEE OCEANS, St Johns, Sept. 2014
viii
Table of Contents
STATEMENT OF ORGINALITY II
DEDICATION III
ACKNOWLEDGEMENTS IV
ABSTRACT V
ASSOCIATED PUBLICATIONS VII
TABLE OF CONTENT VIII
LIST OF TABLES XIV
LIST OF FIGURES XV
ABBREVIATIONS AND SYMBOLS XIX
CHAPTER 1 INTRODUCTION
1.1 BACKGROUND 1
1.2 OBJECTIVES 1
1.3 WHY ACOUSTICS? 2
1.4 COMMUNICATION UNDERWATER 4
1.5 SPATIO-TEMPORAL OCEAN SENSING 5
1.6 RESEARCH CONTRIBUTIONS 6
1.7 ORGANISATION OF THE THESIS 7
1.8 CONCLUSION 8
CHAPTER 2 COMMUNICATION CHALLENGES IN UNDERWATER SWARM SENSOR NETWORK (USSN)
2.1 INTRODUCTION 9
2.2 CHALLENGES IN UNDERWATER WIRELESS SENSOR NETWORKS (UWSN) 10
2.3 TAXONOMY OF MOBILE UNDERWATER WIRELESS SENSOR NETWORK 13
2.3.1 Definition Of Swarming And Robotic Swarm Networks 16
2.4 COMMUNICATION WITHIN UNDERWATER SWARM SENSOR NETWORK 17
ix
2.4.1 Biologically Inspired Underwater Communication 17
2.4.2 Explicit Short-Range Acoustic Communication 19
2.5 UNDERWATER SWARM SENSOR NETWORK (USSN) APPLICATIONS 20
2.5.1 Time Critical Mission Deployment 21
2.5.2 Non-Time Critical Mission Deployment 22
2.6 USSN COMMUNICATION TRAFFIC REQUIREMENTS 23
2.6.1 Payload Data 24
2.6.2 Localisation 24 2.6.2.1 Underwater Vehicle (AUV) Navigation 26
2.6.3 Mission Time Critical: Bio-Inspired Formation Control Algorithms 27
2.6.4 Mission Non-Time Critical: Patterned Formation Control Algorithms 29
2.7 USSN COMMUNICATION CHALLENGES 29
2.7.1 Swarm Network Characteristics 30
2.7.2 Summary Of USSN Data Traffic Requirements 30
2.7.3 Research Goal And Specific Research Questions 31
2.8 CONCLUSION 31
CHAPTER 3 SHORT-RANGE UNDERWATER ACOUSTIC COMMUNICATION CHANNEL CHARACTERISTICS 33
3.1 INTRODUCTION 33
3.2 ACOUSTIC CHANNEL 33
3.3 UNDERWATER ACOUSTIC CHANNEL CHARACTERISTICS 35
3.3.1 Acoustic Signal Level 35
3.3.2 Transmitter (Projector) Signal Intensity 36
3.3.3 Signal Attenuation 36 3.3.3.1 Spreading Loss 36 3.3.3.2 Absorption Loss 37 3.3.3.3 Propagation Loss 39 3.3.3.4 Speed of Sound 40
3.3.4 Underwater Multipath Characteristics 41 3.3.4.1 Swarm Reverberation 42 3.3.4.2 Reverberation 44 3.3.4.3 Reverberation and Transmitter Power Levels 45 3.3.4.4 Delay Spread and Coherence Times 45
x
3.3.5 The Doppler Effect 46
3.4 NOISE 46
3.4.1 Ambient Noise 47
3.4.2 Self Noise 48
3.4.3 Intermittent Sources of Noise 49
3.5 SHORT-RANGE ACOUSTIC COMMUNICATION PHYSICAL LAYER PARAMETERS 50
3.5.1 Signal-to-Noise Ratio 50
3.5.2 Frequency Dependent Component of SNR 50
3.5.3 Channel Bandwidth 53
3.5.4 Theoretical Channel Capacity 54
3.5.5 Receiver (Hydrophone) Signal Intensity 55
3.5.6 Signal-to-Noise+Interference- Ratio (SNIR) 56
3.5.7 Modulation and Bit Error Rate (BER) 57 3.5.7.1 Currently Available Acoustic Modem Capacities 59
3.5.8 Long-Range Vs Short Range 59
3.6 CONCLUSION 61
CHAPTER 4 MEDIUM ACCESS CHALLENGES FOR UNDERWATER SWARM SENSOR NETWORKS 63
4.1 INTRODUCTION 63
4.2 MAC PROTOCOL OVERVIEW 64
4.2.1 Random Access 65
4.2.2 Scheduled Protocols 70
4.3 TIME SCHEDULED MEDIUM ACCESS AND TOKEN POLLING APPROACHES 71
4.3.1 TDMA Based Protocols For Swarming AUVs 71
4.3.2 Token Polling Protocols For Swarming AUVs 73
4.4 CHALLENGES AND OPPORTUNITIES USING TDMA AND POLLING ALGORITHMS 74
4.4.1 Time Synchronisation 74
4.4.2 Guard Time 75
xi
4.4.3 Scalability 75
4.4.4 Time-Slot Scheduling 76
4.4.5 Spatial-Temporal Diversity 77
4.4.6 Application Of Spatial-Temporal Diversity 80
4.4.7 Summary 81
4.4 CONCLUSION 81
CHAPTER 5 INTRODUCTION AND ANALYSIS OF TWO NEW MAC PROTOCOLS FOR UNDERWATER SWARM SENSOR NETWORK (USSN) APPLICATIONS 83
5.1 INTRODUCTION 83
5.2 APPLICATION DEVELOPMENT STRATEGIES 84
5.2.1 Non-Time Critical Mission Deployment 85 5.2.1.1 Non-Time Critical Mission Data Traffic 86
5.2.2 Time Critical Mission Deployment 87 5.2.2.1 Time Critical Mission Data Traffic 87
5.3 ADAPTIVE TOKEN POLLING (ATP-MAC) PROTOCOL DESCRIPTION 87
5.3.1 ATP-MAC Packet Structures 89
5.3.2 ATP-MAC Cycle Description 90
5.3.3 Cycle Time (Tcycle) Analysis 92
5.4 ADAPTIVE SPACE TIME – TDMA (AST-TDMA) PROTOCOL DESCRIPTION 93
5.4.1 AST-TDMA Packet Structure 94
5.4.2 AST-TDMA Cycle Description 95
5.4.3 Cycle Time (Tcycle) Analysis 96
5.5 USING SPATIAL-TEMPORAL DIVERSITY 96
5.6 CONVENTIONAL TDMA PROTOCOL 99
5.7 PERFORMANCE CRITERIA 99
5.7.1 Network Delay 100
5.7.2 Channel Resource Utilisation and Throughput 100
5.7.3 Swarm Synchronisation 100
5.7.4 Performance Boundaries 102 5.7.4.1 NCCPsoft Bounds – Due To Failure 102 5.7.4.2 NCCPhard Bounds – Due To Vehicle Collision 103
xii
5.7.4.3 NCCPhard Bound For Bus Topology 103 5.7.4.4 NCCPhard Bound For Cluster Topology 105 5.7.4.5 NCCP Limits 105
5.8 QUEUING MODEL ANALYSIS 106
5.8.1 Model Parameters 108
5.9 IMPACT OF NETWORK DELAY ON SWARM SIZE 108
5.9.1 Expected Data Packets Per Cycle 108
5.9.2 Cycle Time And Network Saturation 112
5.9.3 Neighbourhood Communication Cycle Period (NCCP) 113
5.9.4 Minimum Packet Arrival Rates 115
5.9.5 Determination Of Maximum Swarm Size 117 5.9.5.1 Variations Due To Packet Length 120 5.9.5.2 Variations Due To Range Between Sequence Vehicles 123
5.9 CONCLUSION 126
CHAPTER 6 AST-TDMA PROTOCOL SIMULATION ANALYSIS AND EVALUATION IN NON-IDEAL UNDERWATER ENVIRONMENTS 127
6.1 INTRODUCTION 127
6.2 SIMULATION MODEL AND METHODOLOGY 127
6.2.1 Modelling of an Acoustic Underwater Channel & Physical Layer in OpNet 128
6.2.2 OpNet Model 130
6.2.3 OpNet Parameters 130
6.3 VALIDATION OF SIMULATION MODEL 131
6.3.1 Protocol Process Evaluation 132
6.3.2 Validation of Simulation Model Results 135
6.4 PROTOCOL MODIFICATIONS 137
6.4.1 Protocol Procedures 137
6.4.2 Additional Protocol Performance Metrics 139 6.4.2.1 Throughput 139 6.4.2.2 Channel Capacity Utilisation 140
6.5 ANALYSIS OF PROTOCOL VARIATIONS 141
6.5.1 Transmission ‘WAIT’ Modification 141
xiii
6.5.2 Packet Size Variations 145 6.5.2.1 Option of Data Packet Train 146 6.5.2.2 Option of Piggybacking of Data Packets 148
6.6 RESULTS IN NON-IDEAL UNDERWATER ENVIRONMENTS 148
6.6.1 The Non-Ideal Channel in OpNet 148
6.6.2 Variations in Transmitter Power 150
6.6.3 Comparison with TDMA protocol and Channel Utilisation Benefits 153
6.6.4 Variations in Packet Length 154
6.6.5 Introduction of Swarm Reverberation 155 6.6.5.1 Noise and Reverberation Levels 156 6.6.5.2 Variations in Transmitter Power 157
6.7 CONCLUSION 158
CHAPTER 7 CONCLUSION 161
7.1 INTRODUCTION 161
7.2 RESEARCH CONTRIBUTIONS 162
7.2 FUTURE RESEARCH 163
REFERENCES 165
APPENDICES 175
APPENDIX A – FISHER & SIMMONS COEFFICIENTS 175
APPENDIX B – MATLAB CODE 176
APPENDIX C – BPSK MODULATION CURVE 178
APPENDIX D – ENERGY CONSUMPTION IN AN AUV 179
APPENDIX E – PROCESS MODEL OPNET CODE 180
xiv
List of Tables
Chapter 1
Table 1.1: Attenuation Comparison 3
Chapter 2
Table 2.1: Single Cluster Underwater Swarm Sensor Network (USSN) Applications 21
Table 2.2: Example of Payload Sensor Types For Mission Time Critical and Non-time Critical Applications 25
Table 2.3: Single Cluster Underwater Swarm Sensor Network (USSN) Traffic Characteristics 30
Chapter 3
Table 3.1: Packet Timing Diagram with Swarm Scattering Reflections for a 4-vehicle Swarm at 30m for different packet sizes 43
Table 3.2: Comparison of Terrestrial and Long and Short range Acoustic Bandwidths 60
Chapter 5
Table 5.1: Application Specific Deployment and Communication Requirement Overview 85
Table 5.2: ATP-MAC and AST-TDMA Packet Structures (bytes) 90
Table 5.3: NCCPhard for vehicle collisions based on Disturbances in Bus 102
Table 5.4: Summary of NCCPhard and NCCPsoft bound for Bus Topology 104
Table 5.5: Hard and Soft Time Boundaries of NCCP for Cluster Topology 106
Table 5.6: Summary of NCCPlimit (s) 106
Table 5.7: Base Parameters used in Initial Analysis 109
Table 5.8: Maximum Number of Vehicles that can be supported in Small Disturbance Model at 50m 119
Table 5.9: Packet Size Determination 121
Chapter 6
Table 6.1: Modified Pipeline Stages 128
Table 6.2: Main Parameters and Transmission Characteristics used in OpNet 132
xv
List of Figures
Chapter 1
Figure 1.1: AUV Swarm, Stylised SeaVision™ Vehicles 3
Chapter 2
Figure 2.1: Taxonomy for Mobile Underwater Wireless Sensor Networks 14
Figure 2.2: Various AUV’s 18
Figure 2.3: Fully Distributed Architecture for a Time Critical Mission using Underwater Swarm Sensor Network 22
Figure 2.4: Decentralised Hierarchical Architecture for a Non-time Critical Mission using an Underwater Swarm Sensor Network 23
Chapter 3
Figure 3.1: Underwater Acoustic Environment 34
Figure 3.2: Block Diagram of a Projector and Hydrophone 35
Figure 3.3: Absorption Coefficient vs Frequency 38
Figure 3.4: Path Loss vs Range 39
Figure 3.5: Typical Sound Speed Profile in the Ocean 41
Figure 3.6: Data Transmission and Swarm Reverberation from a 4 vehicle USSN 43
Figure 3.7: Power Spectral density of the Ambient Noise; W (wind), S (shipping) 47
Figure 3.8: Frequency Dependent Component of Narrowband SNR 51
Figure 3.9: Optimum Signal Frequency based on Optimising SNR (determined from frequency-dependent component of narrowband SNR) 52
Figure 3.10: Range dependent 3dB Channel Bandwidth shown as dashed lines. The Y-axis is the Optimum SNR based on the frequency dependent component of the narrowband SNR 54
Figure 3.11: Theoretical Limit of Channel Capacity (kbps) verse Range 55
Figure 3.12: Receiver Signal Intensity vs Range for Variation in Transmitter Power and Transducer Efficiency 56
Figure 3.13: SNIR vs Range for variation in Transmitter Power, Transducer efficiency, and Reverberation Level 57
Figure 3.14: BER vs Range for Short Range Acoustic Data Transmission Underwater 58
Chapter 4
Figure 4.1: Hidden and Expose Node Problem 66
xvi
Figure 4.2: Minimum CSMA cycle with handshaking 67
Figure 4.3: Spatial-Temporal Diversity 78
Figure 4.4: One Data Exchange Cycle between 2 Nodes for Different β 79
Chapter 5
Figure 5.1: Bus Topology for a Non-time Critical Mission using an Underwater Swarm Sensor Network 85
Figure 5.2: Cluster Topology for a Time Critical Mission using Underwater Swarm Sensor Network 86
Figure 5.3: ATP-MAC and AST-TDMA Protocol Operation showing one full cycle of transmission 91
Figure 5.4: Spatial-Temporal Diversity Explained. A Simple Four Vehicle Topology 97
Figure 5.5: AST-TDMA: One cycle of slot times based on configuration of Figure 5.4 97
Figure 5.6: Determining validity of non-exclusive access 98
Figure 5.7: Potential Disturbance in Bus Topology 101
Figure 5.8: Potential Disturbance in Cluster Topology 104
Figure 5.9: Average Expected Number of Packets Serviced per Cycle for increasing Packet Arrival Rate at 50 m. Comparison of the TDMA, ATP-MAC and AST-TDMA protocols and 5 or 15 Vehicle Swarm 110
Figure 5.10 Packets available in each vehicle per cycle at various Packet Arrival Rates in a 5-Vehicle Swarm at 50 m 111
Figure 5.11: Comparison of Cycle Time, for the three protocols with a 5-Vehicle and 15-Vehicle Swarm at 50m 112
Figure 5.12: AST-TDMA protocol: NCCP, Tcycle and Packet Discards for 5-Vehicle Swarm at 50 m 114
Figure 5.13: ATP-MAC protocol: NCCP, Tcycle and Packet Discards for 5-Vehicle Swarm at 50m 114
Figure 5.14: TDMA protocol: NCCP, Tcycle and Packet Discards for 5-Vehicle Swarm at 50 m 114
Figure 5.15: Comparison of Minimum Packet Arrival Rate for Increasing Swarm size at 50 m 115
Figure 5.16: AST-TDMA 5-Vehicle Swarm at 50 m showing Number of Cycles per NCCP and between packet discarded 116
Figure 5.17: AST-TDMA 5-Vehicle Swarm at 50 m showing Number of Packets queued and discarded 116
Figure 5.18: Determining limit to the Number of Swarm Vehicles using Bus Topology at 50m 118
xvii
Figure 5.19: Determining limit to the Number of Swarm Vehicles using Cluster Topology at 50m 118
Figure 5.20: AST-TDMA NCCP, NCCPlimit and various Packet Arrival Rates 120
Figure 5.21: Maximum Number of Swarm Vehicles in Bus Topology with Changes in Packet Size at 50 m 122
Figure 5.22: Maximum Number of Swarm Vehicles in Cluster Topology with Changes in Packet Size at 50 m 122
Figure 5.23: Maximum Number of Swarm Vehicles in Bus Topology for increasing range 125
Figure 5.24: Maximum Number of Swarm Vehicles in Cluster Topology for increasing range 125
Chapter 6
Figure 6.1 Radio Transceiver Pipeline Execution for One Transmission 129
Figure 6.2a: 5-Vehicle Cluster Topology. Each vehicle represented 130
Figure 6.2b: OpNet Model 130
Figure 6.3: OpNet AST-TDMA Process Model 131
Figure 6.4: AST-TDMA Protocol, 5-Vehicle Swarm of Figure 6.1 (a), illustrating Packet Tx & Rx in each vehicle 133
Figure 6.5: Comparison of Timing between AST-TDMA & TDMA, for V5 of 5 133
Figure 6.6: 5 and 15-Vehicle Cluster Topology Swarm, @ 50 m, initial positions 134
Figure 6.7: Comparison of Cycle Time (Tcycle) obtained in OpNet and MATLAB for both the AST-TDMA and TDMA protocols (compare with Figure 5.11) 135
Figure 6.8: AST-TDMA protocol showing relationship between Tcycle, NCCP and Packet discard. Comparison of OpNet and MATLAB results (compare with Figure 5.12) 136
Figure 6.9: TDMA protocol showing relationship between Tcycle, NCCP and Packet discard. Comparison of OpNet and MATLAB results (compare with Figure 5.14) 136
Figure 6.10: AST-TDMA considerations in protocol operations 138
Figure 6.11: Comparison of Protocols and Number of Vehicles in Swarm at 50 m 140
Figure 6.12: Process Model for Wait Modification 143
Figure 6.13: Comparison of Average Number of Cycles per NCCP 144
Figure 6.14 Comparison of NCCP times between protocols and Tcycle for the AST-TDMA with Token 144
Figure 6.15 Comparison of the true Channel Utilisation Uitrue 144
Figure 6.16 Packet Discards per cycle: Comparison between protocols for a 15 and 5 vehicle swarm at 50 m 146
Figure 6.17 Comparison of NCCP for 15 vehicle swarm at 50 m with various Data Packet Sizes defined in Table 5.9 147
xviii
Figure 6.18 Channel Utilisation at λsat: Comparing AST-TDMA with Wait and TDMA protocol 148
Figure 6.19 SINR for 5-Vehicle Swarm at 20 m 149
Figure 6.20 Average Packet loss for 5-Vehicle Swarm against SINR and Reverberation Levels at 50 m 151
Figure 6.21 Average NCCP for 5-Vehicle Swarm at 50 m showing the increasing NCCP as Packet Arrival rate falls below cycle saturation and variations in Reverberation Levels (Sea States) 152
Figure 6.22 Average NCCP at λsat for changes in SNIR due to Reverberation Levels: Comparison between AST-TDMA and TDMA protocols for 15-V swarm 153
Figure 6.23 Channel Capacity Utilisation at λsat for changes in SNIR due to Reverberation Levels: Comparison between AST-TDMA and TDMA for a 15-V swarm 154
Figure 6.24 NCCP for variations in Data Packet Size for 5-V Swarm at 50 m 155
Figure 6.25: Noise and Reverberation Levels 156
Figure 6.26: Noise and Reverberation Levels for Variation in Transmitter Power 157
Figure 6.27: Packet Loss with variations in Packet Size and Transmitter Power 158
Figure 6.28: Packet Loss with variations in Transmitter Power and Sea State 158
xix
Abbreviations and Symbols
Abbreviations AUV Autonomous Underwater Vehicle ATP-MAC Adaptive Token Polling - MAC AST-TDMA Adaptive Space Time - TDMA LIDCA Lowest Identifier Clustering Algorithm NCCP Time of one completed cycle of exchange of data or Neighbourhood Communication
Cycle Period (everyone-to-everyone) SNIR Signal to Noise & Interference Ratio SNR Signal to Noise Ratio TDMA Time Division Multiple Access QoS Quality of Service USSN Underwater Swarm Sensor Network
Commonly Used Symbols α absorption coefficient B Bandwidth (Hz) Bc Channel Bandwidth c Speed of Sound (1500 m/s) Cc Channel Capacity C Number of cycles per NCCP d Depth (m) Di Number of NCCP cycles in Vi in Tsim Δfd Doppler shift Dpkt������ Expected data packet delay d Distance of Travel of a vehicle DItx Directivity Index Fi Throughput of swarm Formation data in ith Vehicle fmax Maximum frequency fmin Minimum frequency f Carrier frequency (fo is Optimum Signal Frequency) FL Interference Level Gi Throughput of ith Vehicle λ Data arrival rate in a vehicle Lpoll Length of a poll packet Ldata Length of a data packet Ltoken Length of a token packet I Source intensity Iref Reference Intensity Iomni Intensity of spherical spreading Idir Intensity along the axis of the beam pattern
xx
M Maneuverability range (m) ηtx ηrx Projector and Hydrophone efficiency Nq Expected number of packets in a vehicles queue NCCPsoft Complete packet exchange limit based on maneuverability of the vehicles NCCPhard Complete packet exchange limit based on collision avoidance of vehicles NCCPlimit Complete packet exchange limit taking into account NCCPsoft and NCCPhard Nd Average Number of Packets per Cycle Nwind Noise spectrum density from wind Nship Noise spectrum density from shipping Nturb Noise spectrum density from turbulence Nthermal Noise spectrum density from underwater thermal noise Nemf Noise spectrum density from electronic thermal noise Noise Total noise P Pressure Pi Number of Packets successfully received by Vehicle i Ptx total acoustic power consumed by the Projector Paref Reference pressure level , 1 μPa ρ density of the medium (averages for sea water are: ρ = 1025 kg/m3) PLspreading Propagation Loss from Spreading loss PLabsorption Propagation Loss from absorption PLloss Combined Propagation Loss R Packet transmission rate (bps) r Range between vehicles (m) r12 Range between Vehicle with V1 and Vehicle with V2 RL Reverberation Level s Shipping activity factor S Speed of Vehicles S^ Speed of Vehicle with external force added ΔS Relative velocity between moving vehicles SPLprojector Projector source pressure level Sal Salinity t Temperature T Absolute temperature Tcycle Time of one cycle or Vehicle Sequence Time (once through each sequenced vehicle,
whether they sent data or not) tqueue Average queue waiting time tLS Propagation delay between lead vehicle and swarm vehicle tSS Propagation delay between two swarm vehicles tij Propagation delay from Vehicle i to Vehicle j tprop������� Average propagation delay of packets received in a vehicle in one cycle Tslot Slot Size (s) for TDMA protocol Tcomm Transmission Time of a poll packet Tdata Transmission Time of a data packet Ttoken Transmission Time of a token packet Td-t Transmission Time of the data portion of a data packet TX Generalised Transmission Time Tcycle Time to complete a cycle through sequence of vehicles tprocess Processing time required for a packet in the transceiver tcreate Processing time required to create Command packet at start of cycle tcoll Time to vehicle collision θ Angle of disturbance of a vehicle from its planned trajectory
xxi
Tsim Time over which simulation is conducted Titx Transmission time of a packet from ith Vehicle Tirx Reception time of a packet in the ith Vehicle Ui Channel Utilisation V Number of Vehicles in a Swarm Vi Vehicle with ID i w Wind State m/s
1
Chapter 1 Introduction
1.1 Background Mobile swarms of autonomous underwater vehicles (AUVs) have exciting potential for extending
the current operational applications underwater and to add new opportunities to the working
environment of the oceans. Applications include areas such as mapping and surveying [122,
143], military tasks such as to replace workers for dangerous tasks in ocean war zones [31], 3D
plume identification and analysis [104] and other more general scientific and commercial studies
of dynamic oceanographic phenomena such as phytoplankton growth or fish migration [62,
65,114]. Current solutions have been built around static sensor networks and single ROVs
(remotely operated vehicles) and single AUVs. The benefits, however, of several vehicles
working together over any single vehicle include greater speed and range of operation,
increased system reliability and higher quality measurements [31, 122]. To achieve these multi-
vehicle system benefits, data communication between vehicles is essential.
A swarm of AUVs can be considered as being composed of typically many simple,
homogeneous and autonomous agents, deployed in a decentralized mobile topology with
communication on a local level for a combined purpose. Swarm behaviour infers a biologically
motivated behaviour that is exhibited by a set of similar kind of animals that are working
together as a collective, such as seen with insects, birds and fish. The communication protocol
for a swarm needs to facilitate ‘awareness’ of other vehicles in a neighbourhood and needs
each vehicle to be able to work autonomously. Swarm formation control algorithms require at a
minimum, to exchange location and trajectory information from all vehicles to all other vehicles
in a neighbourhood in a continuous fashion, so that a group of self propelled AUVs will be able
to operate in a swarm like fashion.
The growth of underwater operations has required data communication between various
heterogeneous underwater and surface based equipment, which are typically sparsely
deployed. Small Autonomous Underwater Vehicles (AUVs) are a more recent addition to the
equipment used in underwater operations. Most AUV development work, however, has
concentrated on the vehicles themselves and their operations as a single unit [41, 60] where
their communication is with other wired or wireless fixed infrastructure. There has been much
less attention given to the development of groups of autonomous vehicles being deployed in an
autonomous swarm.
1.2 Objectives Swarm operations have many benefits: with the ability to scan or ’sense’ a wider area and to
work collaboratively provides the potential to vastly improve the efficiency and effectiveness of
mission operations. Collaboration within the swarm structure will facilitate improved operations
2
by building on the ability to operate as a team that will result in emergent behaviours [17, 92]
that are not exhibited by individual vehicles. Implementation of swarms of vehicles will greatly
improve on the current ability of single vehicles to survey and explore the oceans.
Advances in the development of Autonomous Underwater Vehicles (AUV) (that include being
smaller, low cost and low power) and their potential to work in swarm like configurations,
necessitates the development of effective communication network architectures and protocols
for short-range wireless acoustic underwater communication. This communication is essential to
coordinate operation of the vehicles as well as to transmit data within the swarm to facilitate the
benefits of operating as a team.
It has been observed that the communication within a swarm network can fall into three
categories; Interaction via Environment, Interaction via Sensing and Interaction via
Communication [17]. The former two are implicit communication techniques that use an indirect
measure, from the sensors or data transmissions themselves, to gain information about what
neighbouring vehicles are doing. Interaction via Communication is an explicit communication
where data is exchanged between vehicles. The body of work in this thesis will be presenting an
explicit communication protocol for the purpose of allowing groups of AUVs to exchange each
other’s navigational data so that the group can implement swarming formations.
The main research objectives are to determine:
1. what the operating characteristics of an underwater swarm of AUVs is and how do these
characteristics impact on the design of an effective swarm communication protocol?
2. what the limits are to the number of vehicles that can operate in close proximity to each other
in an underwater swarm, given the ability to explicitly exchange inter-vehicle data through an
acoustic communication network? and
3. how can a Medium Access Control (MAC) communication protocol be designed to take into
account the constraints of a short-range underwater acoustic channel?
1.3 Why Acoustics? Acoustics remain today the most widely used form of communication underwater due to its
ability to send messages over long distances. Sound energy travels more efficiently in water
than air but still relatively slowly at only 1500m/s ± 3% in seawater depending upon
temperature, salinity and pressure [133, 29].1 Optics work at very short range but require clear
water and electro-magnetic waves have high attenuation as shown in Table 1.1.
1 In deep sea channel, sometimes referred to as the SOFAR (Sound Fixing and Ranging) channel, sound is trapped and travels almost horizontally with reduced path losses as has been shown with the effective method that whales use this channel for long distance communication [67].
3
Figure 1.1: AUV Swarm, Stylised SeaVision™ Vehicles
In comparison to RF, the acoustic channel introduces a very high propagation delay, which is
0.67 ms/m (compared to RF of 3.33 ns/m in air). RF, underwater, even at low frequencies
suffers from extreme attenuation due to conductive seawater and high rates of absorption that
has predominately eliminated its use for underwater communications. The slow speed of
propagation of acoustic signals underwater however has a major effect on the performance of a
communication system. This high channel latency effectively means lower reliability due to the
quality of a single-hop link that can change significantly in the order of time required to send and
receive data and the delays in feedback to any changes in channel state information. In
addition, underwater communication channel characteristics change more dynamically than in
terrestrial channels due to its attenuation, noise and thermal profiles [27]. Thus, in terms of the
development of peer-to-peer communication underwater, the latency of acoustic signals
compared with RF in air requires essentially to redesign communication protocols [4].
The underwater environment can be a very noisy environment: including animal noises; wind,
rain and other natural phenomena such as ice cracking and earthquakes; and shipping and
other man made operations in and on the water. Each of these noise contributors operate in
different frequency bands that together build an ambient noise level that is frequency dependent
with noise levels decreasing with increasing frequencies.
Table 1.1: Attenuation Comparison [133]
Type Frequency (kHz) Attenuation dB/km
Sound 30 5
EM Wave 30 7500
4
Taking advantage of the lower noise profile with increasing frequencies needs to be balanced
with the increasing path loss characteristics with increasing signal frequencies. This means
matching signal frequencies to application and environments is required to improve signal
detection. In addition, multipath can impact severely on data reception and is also affected by
application and environment in which the operations occur.
Multipath underwater can be extreme and this also differentiates wireless communication
underwater to that in air, especially in shallow water where boundary reflections on the sea floor
and surface produce a number of significant propagation paths at the receiver. These multiple
signals that have been reflected, scattered or bent will be themselves impacted by the latency of
the channel and delayed in time, more dramatically than in air. Due to the various path lengths
and timing that these additional signals can take, they may create significant Inter Symbol
Interference (ISI) and errors in symbol detection.
There has been little work done on the short-range acoustic channel model as there has not
been the operational demand for these systems. Recently the developments in underwater
acoustic sensor networking (UW-ASN) and the use of multi-hop networking architecture and
data muling operations have generated interest within the research community to develop
shorter range underwater communication systems [41]. As the knowledge of long-range
channel models are well established, the characterising of a short-range channel model will
initially extrapolate this understanding.
1.4 Communication Underwater The underwater acoustic communication channel is recognised as one of the harshest
environments for data communication, with long-range calculations of optimal channel capacity
of less than 50kbps for SNR (Signal-to-Noise Ratio) of 20dB [124] with current modem
capacities of less than 10kbps [137]. Predictability of the channel is very difficult with the
conditions constantly changing due to seasons, weather, and the physical surroundings of sea
floor, depth, salinity and temperature.
The performance of an acoustic communication system underwater is characterised by various
losses that are both range and frequency dependent, background noise that is frequency
dependent and bandwidth and transmitter power that are both range dependent. In general, the
constraints imposed on the performance of a communication system when using an acoustic
channel are the high latency due to the slow speed of the acoustic signal propagation, and the
signal fading properties due to absorption and multipath interferences, particularly due to
reflections off the surface, sea floor and objects in the signal path. High link latency in a
communication network influences the error control techniques, protocol designs and network
throughput. A specific constraint on the performance due to the mobility of AUV swarms is the
Doppler effect resulting from any relative motion between a transmitter and a receiver, including
any natural motion present in the oceans from waves, currents and tides. Because the speed of
sound in water and the speed of AUVs is relatively similar the Doppler effect is very significant
for underwater communication compared with terrestrial systems that use RF.
5
Short-range underwater communication systems have two key advantages over long-range
operations; a lower end-to-end delay and a lower signal attenuation due to range. End-to-end
propagation at 500 m for example is approximately 0.3 sec which is considerable lower than the
2 sec at 3 km but still critical as a design parameter for shorter range underwater MAC
protocols. The lower signal attenuation means that lower power transmitter are required, which
will result in reduced energy consumption, which is critical for AUVs that rely on battery power.
Battery recharge or replacement during a mission is difficult and costly. The dynamics
associated with attenuation also changes at short range where the spreading component
dominates over the absorption component, which means less dependency on temperature,
salinity and depth (pressure). This also signifies less emphasis on frequency as the frequency
dependent part of attenuation is in the absorption component and thus will allow the use of
higher signal frequencies and higher bandwidths at short ranges. This potential needs to be
exploited to significantly improve the performance of an underwater swarm network
communication system.
A significant challenge for data transmission underwater is multipath fading. The effect of
multipath fading depends on channel geometry and the presence of various objects in the
propagation channel. Multipath occur due to reflections (predominately in shallow water),
refractions and acoustic ducting (deep water channels), which create a number of additional
propagation paths, and depending on their relative strengths and delay values can impact on
the error rates at the receiver. The bit error is generated as a result of inter-symbol interference
(ISI) caused by these multipath signals. For very short-range single transmitter-receiver
systems, there could be some minimisation of multipath signals [55, 136]. For swarm
operations, however, there is potentially a different mix of multipath signals that need to be
considered, in particular, those generated due to the other vehicles in the swarm.
Careful consideration of the physical layer parameters and their appropriate design will help
maximise the advantages of a short range communications system that needs to utilise the
limited resources available in an underwater acoustic networking environment. For the medium
access layer design the unique spatial-temporal characteristics underwater due to the very slow
propagation of sound and low bandwidths available creates a very different set of constraints,
compare to RF, that also need to be incorporated in any protocol design. This is why it is not
straight forwarding in adapting RF solutions to the underwater case.
1.5 Spatio-temporal ocean sensing The shorter ranges expected between vehicles in a swarm topology, means that propagation
delays will be smaller than for the more typical longer-range underwater applications, however,
still significant compared to RF, where propagation delay is considered negligible. In fact, the
transmission time of packets are in the same order of magnitude as the propagation delay,
which creates a unique spatial-temporal environment for underwater communication, and is far
different from what is experienced in a terrestrial RF setting. Exclusive channel access based
on transmission time of data becomes ineffective way to avoid collisions, unless large guard
6
times are incorporated to take into account propagation delays between all possible vehicles in
the network. Therefore, non-exclusive access can occur due to the space diversity, which
allows more than one transmission-reception activity in the channel at the same time.
1.6 Research Contributions The key problem being addressed in this thesis is the medium access control (MAC) protocol for
real time communication in a fully connected but distributed group of underwater autonomous
vehicles (AUV) operating as a underwater swarm sensor network (USSN). USSN will be a game
changer for underwater operations, as it will provide a low cost autonomous search and survey
method for the virtually unexplored vast oceans. The research field of USSN is still in its infancy;
in terms of the vehicles’ design and development; the classification of application areas; and
their traffic requirements as well as the communication protocols needed for swarm operations.
The key contributions of this thesis thus include;
• The development of a short-range underwater acoustic communication channel model
in which the design, development and performance analyse of underwater
communication protocols can be advanced. Specifically the development of a SNIR
(Signal to Noise + Interference Ratio) where the interference due to reverberation levels
caused by the impact of long data packets being sent via omni-directional antennas and
their reflections off the many vehicles operating at close range.
The short-range acoustic channel characteristics are compared to the more traditionally
used longer-range channels and the use of RF in terrestrial environments. This new
short-range acoustic model was implemented in OpNet Modeler®, a sophisticated
communication networking simulation environment, used for the evaluation of the
proposed new MAC layer protocols. This required the modifications to the Radio
Pipeline Stages so that the non-ideal short-range acoustic channel could be executed.
• A proposal for a new type of reverberation; Swarm Reverberation will be shown to play
an important role in the reverberation levels for an USSN. With the application of an
underwater swarming network, which has many vehicles in a dense topology, there will
be an impact on the reverberation channel geometry due to the vehicles themselves
being ‘sound reflective’ objects. This channel geometry together with packet size
creates a unique relationship between range (propagation time) and packet length
(transmission time) which will be shown to impact on the level of swarm reverberation.
• The development of a new Taxonomy for Mobile Underwater Wireless Sensor Networks
based on the network coverage area and density of vehicles required. Underwater
Swarm Sensor Networks (USSN) is thus classified based on their potential deployment
arrangements. USSN are defined as a fully connected and decentralised topology of
numerous vehicles working together collectively with continuous point-to-mulitpoint links
and can operate as a single cluster or multi-cluster. This work contributes to the
7
development of a single cluster swarm either operating in a Bus, requiring a Pattern
Formation Algorithm or a Cluster using Bio-inspired swarm formation algorithms.
These deployment options lead to the development of traffic models and the Quality of
Service (QoS) requirements for a USSN.
• The design and development of two new MAC layer communication protocols that
utilise the unique spatial-temporal environment and the challenging acoustic channel
characteristics underwater for the Bus and Cluster topology deployments.
For the Bus Topology, an Adaptive Token Polling protocol, ATP-MAC, uses a polling
approach in a decentralised hierarchical topology. A revised design of the Token Polling
protocol was develop for a decentralised distributed MAC protocol, Adaptive Space
Time – Time Division Multiple Access (AST-TDMA). Both protocols are designed to
effectively use a single channel broadcast acoustic environment while incorporating a
method to handle the spatial-temporal characteristics experienced underwater. They
are both designed to work independently of time synchronization and require no prior
knowledge of propagation delays and future knowledge of the swarm network topology.
An analytical framework using a queuing model to evaluate the performance of the two
adaptive protocols was completed. This found that under ideal underwater channel
conditions and fixed data rates there is a trade-off between range, data packet size,
number of nodes in a cluster and the arrival rates of data in each vehicle queue to
maintain an average packet transfer delay.
• Two new performance metric are developed for analyses of the protocol: NCCP,
Neighbourhood Communication Cycle Period, that establishes the delay in the
distribution of one cycle of navigational data throughout the swarm; and Channel
Capacity Utilisation that investigates the extent to which the channel is utilised which
needs to be maximised for underwater use, as it cannot be used for anything else.
1.7 Organisation of the Thesis This thesis presents the work on the background and development of underwater
communication protocols for underwater swarm sensor networks. Chapter 2 presents a new
taxonomy proposed to classify Underwater Swarm Sensor Networks and the design challenges
and objectives for an explicit communication network between the vehicles are discussed.
Chapter 3 investigates the short-range underwater channel characteristics and explores the
benefits and limitations that this environment imposes on the development of a communication
protocol. A short-range acoustic channel model is developed for the design, simulation and
analysis of the new protocols. In Chapter 4, the state of the art of medium access control (MAC)
layer protocols for Underwater Wireless Sensor Networks is presented with the arguments for
the approach taken in the new protocol designs. Chapter 5 presents the two new protocols and
the analytical analysis of them under ideal conditions while Chapter 6 presents the simulations
in a non-ideal channel.
8
1.8 Conclusion The aim of this work is to develop a communication protocol for a swarm of AUV’s. This
communication protocol requires that each vehicle’s location and navigational information is
exchanged with all other vehicles in its cluster, so that it can operate in a swarm-like behaviour.
The purpose is for future systems to be able to build on the benefits of cooperation between
vehicles and to perform collaborative missions. For this to be achieved there may be additional
information or sensor data required to be exchanged.
Creating underwater swarm sensor networks poses many new challenges for researchers, due
to the unique communication environment that exists, which has meant that many of the
techniques used in RF wireless communication do not apply. A good understanding of the
propagation channel is essential for both the design of and performance evaluation of an
underwater communication network. Due to the high propagation delay of an underwater
channel, any change of link quality such as SNIR will significantly affect the performance of the
network. MAC protocol designs require transmission channel state information in order to
optimise their performance. Hence, it is necessary to develop a new class of protocols which
can adapt themselves with the varying channel conditions and offer reasonable high throughput
in swarm networks.
9
Chapter 2 Communication Challenges in
Underwater Swarm Sensor Network (USSN)
2.1 Introduction The focus of this chapter is to introduce the research and develop associated with
communication requirements in a swarm of underwater vehicles. In particular to:
(1) Review the body of knowledge and current projects and practices associated with the
networking and communication requirements of Underwater Swarm Sensor Networks
(USSN);
(2) Explore the potential application areas to establish the architecture and traffic requirements
of a swarm network; and
(3) Define the design criteria for the explicit communication requirements of an underwater
swarm network.
Research into developing network solutions for terrestrial wireless ad-hoc and sensor networks
(WSN) has been active for many decades with more recent focus also including advancing
aerial swarming sensor networks. Only in recent years have the advances in underwater
technology enabled exciting new opportunities for UWSNs to be implemented to monitor larger
areas of the vastly unexplored oceans.
The growth of wireless systems terrestrially has been pushed by the sudden growth in wireless
communication technologies, which has enabled enormous expansion of new application areas
such as military, habitat and environment monitoring and observation as well as aerial swarm
networking. Similarly, there are significant advances in research and operational development
of underwater wireless network structures that focus on fixed infrastructure. It is envisioned
however that mobile groups of UWSNs, or USSN, will become very important because of their
relative ease of deployment, absence of cables, and their ability to adaptively sense a large
area. Despite this, autonomous swarms of mobile vehicles and even the integration of mobile
vehicles into fixed sensor networks are still a unique and growing area of research interest [13].
The field of Underwater Swarm Sensor Networking (USSN) requires the combining of the two
fields of formation (swarming) algorithms and underwater mobile WSN technology. The area of
swarming algorithms continues to advance and can be directly adapted from terrestrial to
underwater environments. It is however the developments of underwater mobile WSN
technologies that still face many challenges which are substantially different to a terrestrial
setting.
Swarming, infers a biological process, and thus the swarming algorithms are predominately
being developed from bio-inspired swarming formation control processes and systems.
10
Adaptation of these algorithms to underwater applications will be considered here with
emphasis on their impact on the communication protocol requirements. An examination of the
factors that influence the design of the network topology and data communication requirements
for an underwater swarm network will thus be reviewed.
The WSN developments for underwater mobile applications, still face many challenges primarily
due to the resource constraints imposed by the underwater environment that are far more
limiting than in more traditional wired and terrestrial wireless environments. With the progress
and growth now occurring in underwater technologies, and recently with Autonomous
Underwater Vehicles (AUVs), the underwater world of mobile sensor networking is set to
expand [56]. This requires wireless communication between vehicles.
The age of USSN is thus beginning and with it the need for short-range underwater
communication and swarm communication networking protocols. Traditionally communication
underwater is by an acoustic medium, rather than electromagnetic, and this brings with it the
requirement for new approaches in networking and communication design [4, 32].
This chapter will begin with an overview of the major difficulties that underwater wireless sensor
networks face including a discussion related to the differences between terrestrial and
underwater operations. A taxonomy for Mobile UWSN was developed to put a context to the
discussion that will follow which will focus on the communication requirements of an Underwater
Swarm Sensor Networks (USSNs) that forms one part of the field of Mobile UWSN. This
taxonomy is used to discuss the communication challenges that these networks need to
overcome. A literature review of the few projects that have examined this specific area will be
discussed, however the literature related to the broader area of acoustic communication in an
underwater wireless sensor network (UWSN) will also be reviewed to present a broader
background to some of the potential interesting developments that need to be considered. A
discussion on swarm algorithms provides a view on the traffic requirements and network
structures that may results. A summary of the communication challenges that will be
investigated in this work will then be presented.
2.2 Challenges in Underwater Wireless Sensor Networks (UWSN) Operating communication networks underwater is substantially different to terrestrial and space
operations and thus we begin this work with a list of the major challenges and principal
differences between terrestrial and UWSN. This list is divided into primary and secondary
issues. Primary issues focus specifically on the design and development needs of an acoustic
communication systems in an underwater wireless network and these will be investigated
further as we review and develop the requirements of an underwater swarm network. The
secondary issues relate to the broader network and technology matters that may indirectly
impact on the design of underwater networks and the limitations that these may impose on the
communications.
11
Primary Issues
• Bandwidth: The underwater acoustic channel is considered one of the most difficult
operating mediums for data communication. Both noise and propagation losses are
frequency dependent and limit the operating frequency and bandwidths to low kHz
[134]. The severely limited bandwidths available underwater have a major impact on
network structures and protocols due to the trade-off between network node densities
and information exchange requirements. Chapter 3 will explore the channel issues in
more detail.
The lower operating frequencies create fundamental physical bandwidth limits. Not only
are the absolute bandwidths low but also they are not negligible with respect to centre
frequency, with bandwidths (B) in the low kHz and centre frequencies (fc) in the low 10’s
kHz (i.e. B is in the order of fc). Thus the generalised narrowband assumptions made in
RF communication of B << fc do not hold underwater [126, 9], and therefore the
assumption that the behaviour across the bandwidth will be the same. This is critical for
signal processing and synchronisation but also implies the need to respect the band-
limited nature of these systems at the MAC layer to develop bandwidth efficient
modulation and protocol solutions.
Underwater acoustic communication is nearly always half-duplex due to the very small
frequency bands available and also for space constrained AUVs, transmitters and
receivers cannot be spatially separated far enough to provide full-duplex connections
[101].
• Latency: Sound underwater travels at approximately 1500 m/s, which is very much
slower than the speed of light (electromagnetic radiation) at 3 x 108 m/s. This means
large propagation delays and can lead to relatively large motion-induced Doppler effects
which can mean even at very short distances high multipath spreads of 10 to 100 ms
can occur [126].
• Power: In underwater acoustic networks the transmit power is typically several
magnitudes higher than the received power [101]. For longer ranges this can be in the
order of up to 100 times while at very short ranges goes down to less than 10 times
[137]. This is very different to most terrestrial applications where the transmitting and
receiving powers are approximately the same.
• Deployment: UWSNs are generally more sparsely deployed and employ considerably
fewer nodes compared to terrestrial WSNs due to the cost of underwater hardware (see
Component Costs below) as well as operational deployment costs [56, 101].
• Duration: Underwater wireless sensor networks are generally deployed over shorter
periods - from several hours to weeks. This is considerably different to terrestrial sensor
12
networks where, depending on the application, it is more common that their deployment
is for several months to years.
• Communication Addressing: Trends towards data centric communication networking is
occurring in terrestrial sensor networks due to the large-scale and dense deployment of
larger systems. As underwater sensor networks use much smaller numbers of units and
are more sparsely deployed, address-centric methods are more practical.
Secondary Issues
• Energy Consumption: the energy consumption requirements for underwater acoustic
sensors are much higher than those required in RF sensors due to the more complex
signal processing capabilities required in the receiver to compensate for the harsh
underwater channel conditions and the higher transmitter power requirements for the
acoustic rather than RF physical layer discussed above [4].
As battery capacity can be limited due to size restrictions on small AUVs and recharging
is virtually impossible underwater, battery power is a limited resource.
• Component Costs: While the cost of terrestrial sensors continues to decrease at a rapid
rate, underwater sensor costs remain high. This is because of the lack of economies of
scale as well as the significantly higher manufacturing costs due to materials and
techniques required to combat the harsh operating conditions experienced in water.
This is expected to change, albeit slowly, as underwater operational work becomes
more commonplace. Both terrestrial and underwater devices have benefited from the
miniaturisation of sensor technologies with smaller chip sizes driving down power
consumption and therefore improving their energy efficiency.
• Economics: There are several significant economic differences between terrestrial and
underwater networks, particularly those that relate to operational aspects of deployment
and recovery as well as to component costs, as discussed above. Launch and recovery
costs for underwater sensor networks are typically much higher due to the need to use
either oceanographic research vessels or commercially operated ships that are often
needed for several days at a time. Much of the time and therefore cost is due to the
difficulty of recovery, which is still considered essential because the devices are too
expensive to be considered disposable. Concerns are being raised with abandoning
items at the end of projects as rubbish, as this means a build-up of litter that is already a
problem in our oceans.
In addition, because the bandwidths are poor, full data recovery is often planned on
retrieval. GPS technology is being considered more for both node recovery and data
downloads but it has its limitations. As radio waves suffer from high attenuation
underwater, GPS cannot function underwater and is only advantageous when vehicles
surface. This functionality would make recovery quicker and easier but it adds costs to
incorporate it on vehicles when it may not play any other part.
13
The infrastructure costs of underwater sensor networks and their deployment
underwater are significantly higher than for terrestrial systems which also has an
economic impact on these systems.
• Environment: Seawater is a particularly harsh environment for most materials which
means underwater sensors are much more prone to failure because of fouling and
corrosion than terrestrial sensors, which do not generally require maintenance.
• Data Storage: The data storage capacity requirements in underwater sensors will
generally be higher than those of terrestrial sensors. This is because of the greater
potential for connectivity losses due to poor channel conditions and the lack of spatial
correlation underwater as GPS or other forms of positioning techniques are not
available.
2.3 Taxonomy of Mobile Underwater Wireless Sensor Network To begin this work and to provide a context in which to place this work, a new taxonomy for
mobile UWSN has been developed, see Figure 2.1, adjusted from the taxonomy for underwater
acoustic networks [101, 74]. This taxonomy for mobile UWSN is build from the possible network
operating environments and is used to discuss the implication of these classifications on the
communication protocol structures. These operating environments are based on the
deployment configurations expressed by the density of vehicles and sensor network coverage
area, which is defined as a volume due to the 3D nature of operations underwater [105]. These
two dimensions impact on the design of the MAC and network-layers [101] and will provide a
framework in which a broader discussion on the communication requirements of an Underwater
Swarm Sensor Network can be defined. In the following section, this will be expanded upon
through an evaluation of the literature available.
Underwater mobile networks that cover large geographical areas, identified by the top two ‘blue’
quadrants in Figure 2.1, have been recognised to have similar network characteristics to those
encountered in Delay / Disruptive Tolerant Networking (DTN) [10, 53, 54, 74, 101, 146]. While
originally developed for deep space networking and interplanetary communication, DTN is a
field that is seeing significant research and development due to its applicability to satellite and
sensor-based networks as well as acoustic and underwater applications [48, 70, 139].
Characteristics of a DTN are that may lack continuous network connectivity, have long and
variable delays, have limitations due to the wireless range, high error rates, asymmetric data
rates and have demanding energy and noise issues [85,139]. A DTN can be a network of
smaller networks or subnets, and generally have to deal with disruptions due to link outages
that are likely to occur due to large distances between mobile nodes as well as orbital
mechanics issues in space or topographical disorientation due to mobility of receivers [138].
14
Decreasing Latency
Increasing Connectivity
Increasing Throughput
Increasing Vehicle Collision Potential
Ra
ng
e
det
er
mi
ne
d
Ac
ou
stic
Ra
ng
e
S
m
ll
La
rg
Networ
k
Covera
ge
Volume
Link Laye
r becomes partitioned.
Hidden/exposed
terminals
Link Laye
r unpartitioned. Hidden/exposed
terminals
is common.
Small Numbers Large Numbers
Density of Vehicles
Figure 2.1: Taxonomy for Mobile Underwater Wireless Sensor Networks
These characteristics and features are analogous to underwater long-range networks, although
the range magnitudes are different (kilometres (underwater) rather than tens of km (space)),
and thus there are several important similarities in terms of communication structures that can
be considered when designing underwater networks. In both environments intermittent
connectivity leads to the absence of an end-to-end path between a source and destination, that
is called network partitioning, and requires specific communication techniques to allow a
network to continue to function. In addition, even if it is not absent, the end-to-end path can
experience significant delays due to the long and variable propagation delays between nodes
and the variable queuing delays that can be expected at a node. In this case, ACK or
retransmission strategies need to be carefully implemented to minimise further delays. With
higher error rates expected there is a need to consider either bit error correction or
retransmission of whole packets which results in more processing and network traffic which will
reduce bandwidth efficiency of a network that is of particular concern in underwater acoustic
communication. To overcome these problems in the terrestrial/space environments DTNs have
successfully adapted a multi-hop ‘store and forward’ message switching approach. This
provides a mechanism where data transmissions are held when a link is unavailable and then
Underwater Mobile DTN
Extremely Sparse
Underwater Mobile DTN
Multi-subnets
Underwater Swarms
USSN Single Cluster
Underwater Swarms
USSN Multi-Cluster
AUVs, Gliders& Drifters Multi-Hop
Point-to-point
Point-to-multipoint Single-Hop
AUVs
15
allows it to continue when a path to its destination becomes available, and therefore a delay will
only occur on one of the links between source and destination.
Multi-hop ‘store and forward’ and other approaches are also being investigated for long range
underwater mobile DTNs. Thus, irrespective of density, communication in these networks are
defined as generally multi-hop and point-to-point. This has been shown to be beneficial in long-
range underwater networks, as it can improve energy efficiency due to the reduction in power
requirements to send data along shorter distances which is a major challenge that these
networks face [2, 19, 148].
In these networks, irrespective of environment, the predominate requirement of the Link Layer is
to maintain fairness among nodes and effective path determination through the network as well
as careful bit error correction and retransmission strategies while the routing layer has to deal
with the extremely long propagation delays. When the number of vehicles is extremely small
and coverage area is very large, the network will reach a limit of overlapping mobile coverage in
which case the network may not be able to form without other infrastructure in place. For the
other extreme, with networks growing large with increasing number of vehicles the current
limiting factor underwater will be the prohibitive costs associated with the devices themselves.
As the network coverage volume decreases and therefore the range between vehicles
decreases, as defined in the bottom two ‘green’ quadrants of Figure 2.1, all vehicles are defined
to be within direct contact of each other, and as such single-hop acoustic communication
networks are possible. Depending on the size of the network, they can operate as a single-
cluster or multi-cluster network depending on the application and traffic requirements. In either
case, these mobile networks have a high vehicle density compared to the DTNs and will be
referred to here as Swarm networks.
Communication between vehicles in these swarm networks can be point-to-point when
networks may only have a few vehicles however in general these networks are more likely to be
point-to-multi-point to increase distribution of information throughout the network. As the range
between vehicles becomes smaller and vehicles are operating within 10s of metres of each
other, vehicle collisions are a serious consideration and navigational information becomes
critical. As the numbers of vehicles in the network increases, so will the data throughput of the
network which the communication protocols and particularly the MAC layer will need to handle.
Maintaining collision-free messaging will support this increase as retransmission and additional
traffic can be avoided.
There has been significantly less research and development work done in the USSN field due to
the focus on applications around long-distant operations and the cost of technology
development and deployment underwater as discussed above. Within the research community,
however, there is a growing interest in underwater swarm sensor networks and it is in this area
that this work will concentrate. More specifically, the focus will be on the single cluster USSN,
represented in the bottom left quadrant of the Taxonomy, which will be more realistically the first
16
development systems primarily due to the cost of AUVs and the fact that they will be less
complex in nature to deploy and test.
Thus a definition and brief description of a robotic swarm will be presented.
2.3.1 Definition of Swarming and Robotic Swarm Networks
Swarms are systems where many individuals are organised and coordinated by principles of
decentralized control, self-organization and at least some form of local communication within
the swarm. There may also be remote communication to a supervisory or control node. A well
known and used definition of swarm (robotics) networks is taken from Sahin [112]: "Swarm
robotics is the study of how large numbers of relatively simple physical embodied agents can be
designed such that a desired collective behaviour emerges from the local interactions among
agents and between the agents and the environment". Sahin [112] also sets out four criteria that
apply when determining the degree to which the term "swarm robotics" should apply in a
specific case:
1. Large numbers of robots: The number of agents must be large or at least the control rules
allow it to be scalable
2. Homogeneous groups of robots: Swarm are often made up of a homogenous group of
agents or if heterogeneous then only with a small number of different types of agents
3. Relatively incapable or inefficient robots: A group of collaborating agents is required
because an individual agent is 'incapable' of completing a task
4. Robots with local sensing and communication capabilities: This ensures that
coordination is distributed [112].
The advantages of swarm networks are that they can cover a large area in detail both in terms
of the static coverage area based on the number of agents and over time with the mobility of the
swarm. Robotic swarms, which will be referred to as swarms, can perform monitoring and
search tasks as well as ‘real-time’ problem solving where they can act to prevent the
consequences of that problem [93]. The autonomy also means that they are very suited to
dangerous tasks, such as searching in mine fields and in dealing with hazardous events like
chemical leaks.
In essence, the swarm network provides the infrastructure that facilitates the collaborative
behaviours being implemented. The first and most fundamental design decision required is
whether the architecture of the network is to be centralised or decentralised, and if
decentralised whether it is hierarchical or distributed. Decentralised networks are claimed to
have several advantages over centralised networks, such as, reliability and scalability, and are
the predominant paradigm discussion and used today [93] as they reflect the 'biologically
inspired' notion of swarming in terms of 'emergent properties' and 'self-organisation'. However,
one of the open research questions is: does the scaling advantage of decentralised networks
offset the coordination advantage of centralised networks?
17
The swarm collaborative behaviour infers a biologically motivated behaviour that is exhibited by
a set of similar kind or size of animals that are working together as a collective, such as swarms
of insects, flocks of birds and schools of fish. As described in Criteria 3 above, at an individual
level each agent can be modelled by a simple set of rules (such as Boids Algorithm described
below) that may not be complex yet the emergent behaviour of the swarm can be quite complex
and harder to model. These rules will need some information from the other agents and thus as
Criteria 4, sets out, some form of communication between agents is essential.
Bio-inspired networking techniques have begun to emerge over the last decades to take a new
approach in some of the most challenging areas of network developments such as large-scale
systems, heterogeneity and unattended operations. These bio-inspired solutions are being
sought from the bio-inspired computing and system application domains and look first at the
identification of analogies with well-researched biological systems. The biological principles of
swarm intelligence and social insects have found equivalencies in several network areas
including distributed search and optimisation, task and resource allocation, and WSNs (Wireless
Sensor Networks) [40].
2.4 Communication within Underwater Swarm Sensor Network The current research and development in underwater swarm sensor networks has focus on
three major aspects:
• AUV design [28, 96, 121, 149];
• Development of the 'bio-inspired' swarm algorithms associated with localisation, formation
and cognition to build the collaborative behaviours [97, 115, 116, 140, 145]; and
• Communication requirements that are recognised as essential for implementing swarm
behaviours [22, 95, 115, 146].
The focus of this work is in the development of communication algorithms and thus the
remainder of this section will investigate the current developments in this area firstly focusing on
research around bio-inspired underwater communication and then more broadly on other short-
range approaches. A review of some of the localisation and formation algorithms for swarming
AUVs will be presented in Section 2.7 as the data requirements will impact on the network
traffic.
2.4.1 Biologically Inspired Underwater Communication
For a group of underwater AUV’s to operate in a swarm like fashion they will need to have both
a swarm formation control algorithm and a communication system which is essential for
neighbouring vehicles to inform each other of formation actions and their location [82]. The
swarm control algorithms discussed in Section 2.7, require knowledge of the location, trajectory
and/or the ‘presence’ of at least one of the closest neighbour vehicles. It has been observed
that the types of communication that are available within swarms can fall into three categories;
Interaction via Environment, Interaction via Sensing and Interaction via Communication [18].
18
(a)MONSUNII UniofLubeck
(b) SHOAL BMT Group (c) CoCoRo Consortium
Figure 2.2: Various AUV’s
Interaction via Environment is an implicit communication system, which is also referred to as a
"cooperation without communication" [6] system. It is based on similar principles to the Ant
Colony Algorithm derived from the behaviour of ants finding food, where they leave 'pheromone
trails' which gain strength when they are the shortest path. The SHOAL project [121] has used
this technique in an interesting ways by leaving markers for navigation and to identify
measurement locations.
The SHOAL project uses four fixed bottom mounted sensors for localisation interacting with
mobile units so it is not a fully mobile autonomous swarm. Each SHOAL vehicle, Figure 2.2 (b)
has incorporated a set of AI (Artificial Intelligence) rules similar to the ACO approach that uses a
‘pheromone’ trail where obstacles are marked with a ‘potential’ and the target as a ‘sink’. The
vehicle uses the potential and sink as forces to navigate the vehicle away from obstacles and
towards the target.
Interaction via sensing is also an implicit communication system. It relates to techniques that
allow local interaction between vehicles using sensing systems that do not involve explicit
communication. The CoCoRo (Collective Cognitive Robots) project [27] is investigating several
types of sensor approaches to detect interactions, including active sonar, optical sensors and
blue LED light. Similarly, the MONSUN II AUV [90], Figure 2.2 (a) has integrated infrared
distance sensors into its front fins to avoid collisions with lateral obstacles, as well as a
visualisation method which uses the camera on the front of the vehicle to allow that vehicle to
follow another vehicle. They also plan to implement a frequency identification system in each
vehicle, based on the system believed to be used by dolphins to identify each other in a pod
[95]. Chen [22] uses the idea of using low-frequency short ‘whistle’-like messages to emulate
the long-haul vocalisation used by killer whales that has been important in the development of
long-distance acoustic communication. He has also incorporated an acoustic echoing system,
19
similar to that used by bats, to obtain an estimate of the Doppler frequency shift of a signal,
which is then used for relative speed calculations.
Interaction via communication is an explicit communication where data is exchanged between
vehicles on a local and/or global level and can include either direct messaging from one
member specifically to another or by broadcast where the recipient maybe either known or
unknown. The focus of the work described in this thesis is on this explicit communication and
how many vehicles can access the wireless medium to exchange the required information.
2.4.2 Explicit Short-Range Acoustic Communication
So far many of the projects that are developing underwater robotic swarms and using explicit
communication techniques are incorporating surfacing vehicles so that they can access GPS
signals for positioning and communication. Most commonly in homogeneous swarms is for each
vehicle, on a rotation basis, to take a turn on the surface [22, 90, 98, 103]. Alternatively, in a
heterogeneous swarm where one type of vehicle is designed as a base station and remains on
the surface throughout a mission [28] or where malfunctioning vehicles float to the surface and
active GPS for recovery [122].
The aim of the MONSUN II project [90] is to be able to deploy an operating swarm of
homogenous vehicles having a distributed hierarchical architecture on long-term operations to
monitor underwater environmental conditions. The approach taken by MONSUN II is to have at
least one of the homogenous vehicles floating on the surface to receive GPS signals that
maintains a fix on the swarms’ absolute position. Submerged AUV's will calculate their position
by the distance to the floating AUV and to their local neighbours using Received Signal Strength
Indication (RSSI) and the transmitted power levels. The vehicles rotated on a regular basis
taking on the role at the surface which includes becoming the lead vehicle until the next vehicle
surfaces. This gives the swarm the advantages of being able to access GPS signals for
absolute positioning, and to balance energy across the network through vehicle rotations.
Chen [22] similarly uses rotating vehicles on the surface for absolute positioning and to use the
last vehicle that surfaced as the lead vehicle as it has the most up-to-date position information.
Using a centralised approach, the new lead vehicle broadcasts a position report message to the
followers. Via a handshaking strategy the followers send position information in return. Once the
leader has received all position information, it can run the formation-mapping algorithm to find
the best position for all swarm vehicles. It broadcasts this back to followers who reply with an
acknowledgement. Chen also uses some implicit communications for relative positioning of
each of the vehicles in an attempt to reduce the communication overheads and this is done via
active sonar, which is a similar approach to bats echolocation mechanism. Here the Doppler
effect allows a vehicle to determine if a neighbouring vehicle is moving towards or away from it.
This worked showed the formation control coordination communication overheads were 2 times
higher than for terrestrial wireless channels due to the less reliable channel as well as the extra
retransmission required due to using contention based MAC layer.
20
The communication networks designed by Petillo [103] and Paley [98] also rely on vehicles
surfacing periodically to access the RF network and to obtain GPS co-ordinates. Petillo
investigates the use of a swarms to monitor and follow an offshore ‘plume’, such as an algal
bloom or an oil spillage by using the RF network to obtain information that could re-direct
individual vehicles to more optimal sampling positions. Petillo believes that until better swarm
communication systems are developed, ‘Periodic Surface Communication' is necessary. The
disadvantage of the approach is that it takes away the ability to maintain a practical, real-time
operation, which limits the application and operational functionality and efficiency of the
network. This includes limiting the depths a swarm can access.
The CoCoRo (Collective Cognitive Robots) project [28] took a different approach for the network
architecture. It allows the swarm to have surface vehicles but also to access the sea-bottom.
CoCoRo incorporates two types of AUV's into a heterogeneous swarm network made up of 3
sub-swarms [116], Figure 2.2 (a). The two types of vehicles are a base station AUV that
accesses GPS and has a recharging facility and the CoCoRo design AUV that is an unusual but
simple U shape platform that can swim in 3-dimensions. As the aim is to develop sea bottom
search capability, the swarm is made up of 3 sub-swarms: the base station floating on the
surface, a group of CoCoRo AUV's used as a relay-swarm to communicate information from the
base station to the 3rd sub-swarm, the sea-bottom-swarm, which searches the sea-bed for its
specified target. The major focus of this work is on the nature of the cognition and 'self
awareness' required by each individual swarm vehicle and many biologically inspired solutions
are being investigated. The project is still at a very early stage of development and
communication protocols between robots are not specifically being investigated.
One of the oldest (known to the author) underwater swarm projects is Serafina [149], which has
since been discontinued, focused on the development of a vehicle that is very similar to the
MONSUN II vehicle, and uncommonly investigated low frequency RF signalling. In this project a
distributed algorithm using time based scheduling was presented. Time-based communication
algorithms for underwater swarms have also been used by several other projects [43, 87, 113,
123] and these will be examined in more detail in Chapter 4. The remainder of this chapter will
establish the application and traffic requirements, and ascertain the communication criteria and
boundaries for this work.
2.5 Underwater Swarm Sensor Network (USSN) Applications The classification of application scenarios for UWSN has regularly been divided into variations
of delay intolerant or delay tolerant data requirements [5, 32, 105]. For USSN we propose a
modification to this classification that focuses on the network architecture and the consequential
required formation control algorithms that best suit the applications such that applications for
swarm networks can roughly be divided into Mission Time Critical or Mission Non-Time Critical
see Table 2.1.
21
Table 2.1: Single Cluster Underwater Swarm Sensor Network (USSN) Applications
Application Category
Mission time critical Mission non-time critical
Applications Search, identification, target (objects, fish,
organisms, resources, pollutants, etc.)
Inspections (harbours, structures, pipes) and
Military (Mine-countermeasures, hazardous
jobs)
Scientific / Environmental sampling and
monitoring (salinity, temperature, sounds,
oxygen levels, hydrothermal helium, fish
migration, currents, pollutants, etc.)
Oceanographic Surveying and
Area Surveillance and Protection
Architecture Multi-vehicle coordination in a swarm like
arrangement travelling in a random ‘bio-inspired’
formation control configuration
Multi-vehicle coordination in a swarm like
arrangement traveling in a structured pattern
Operations Swarm continuously changing shape
Incorporating payload data in real time for
mission completion Monitor for specified periods Predominately Shallow (up to 300m) Offshore, harbours, rivers etc
Analysis and incorporation of real-time payload
data into formation control algorithm
Endurance of several hours to days
Swarm maintains structure Collect and store payload data for later
downloading
Continuous monitoring Shallow to Deep (up to 3000m) Predominately Offshore but also harbour and
estuaries
Endurance of day(s) / month(s)
Data exchange Continuous real-time localisation and payload
data exchange for speedy mission completion
Regular localisation data exchange to
maintenance formation structure
Energy Requirements
Hours of Operation
Day(s) to month(s) of Operation
2.5.1 Time Critical Mission Deployment
Mission Time Critical applications are the more traditionally expected swarming style network
where vehicles operate in a ‘bio-inspired’ formation control pattern and endeavour to gain from
the power of a swarm’s intelligence. Examples of these applications include searching and
finding a target or object such as a black box, a geological vent or a pollutant source such as
finding a chemical leak. Inspection of harbours or underwater structures and military
surveillance applications including mine countermeasures and gathering information in the
battlefield are also generally time critical missions. These missions require speedy discovery
and often-urgent responses and are operating for short durations, lasting for hours rather than
days until targets are found.
The concept of swarm intelligence is important to the network structure and mission approach.
Swarm intelligence describes the behaviours that result from the local interactions of the
individual vehicles with each other and with their environment [58]. There are interesting
emergent properties that occur on the global scale in large swarms even when
22
Figure 2.3: Fully Distributed Architecture for a Time Critical Mission using Underwater Swarm Sensor Network
individuals have a restricted view of the system and only have interactions between neighbours
on a local scale, while operating in a coordinated way without a coordinator or external
controller. Most of the solutions, if appropriately modelled, are built on simple concepts and
rules, which are described in the Section 2.6.
These rules or formation control algorithm together with the payload sensor data that supports
the finding of the target in Time Critical Mission applications defines the data exchange
requirements, which is therefore strongly influenced by the real-time localisation and payload
data collected. This creates a random pattern of movement as vehicles manoeuvre in a swarm
like fashion and is referred to in this work as a Cluster Topology. The Cluster Topology,
illustrated in Figure 2.3, reflects the standard definition of a swarm.
This type of applications demands continuous exchange of real-time localisation and payload
data for speedy mission completion.
2.5.2 Non-Time Critical Mission Deployment
Applications considered as non-time critical missions include environmental and scientific
sampling or surveying for mapping or bathymetry. These applications require that the payload
sensor data is collected along side the location where it was collected and does not need the
same level of real-time interaction or ‘swarm intelligence’ for enabling a speedy mission
completion. The focus in these applications is on the regularity of payload data collection with
the importance on the accurate location where the payload data was collected. This regularity
Surface
Sea Floor
Swarm
Vehicles
23
Figure 2.4: Decentralised Hierarchical Architecture for a Non-time Critical Mission using an Underwater Swarm Sensor Network
suggests a need for a structured pattern of formation where the exchange of data is focused on
maintenance of the structure. As GPS is unavailable underwater there are benefits that multi-
vehicle collaboration can have on better determination of their position [98, 122] especially over
longer missions.
Thus for these applications, vehicle deployment necessitates a structured and stable pattern of
motion that offers a consistent and steady sweep of an area using an arrangement of vehicles
in a line or V pattern that will be referred to in this work as a String Topology. The V pattern
seen in Figure 2.4, which can extend out to a line of vehicles, is used to sweep the widest area
while keeping communication ranges between vehicles as small as possible.
These missions may require days or months of operation and require a regular exchange of
localisation data to maintain accurate location and formation structure. The payload data needs
to be collected in association with the vehicles position and can be stored and retrieved at a
later time when vehicles are recovered.
2.6 USSN Communication Traffic Requirements Sensor data is collected by each AUV for both navigational and mission purposes. Mission data
is the sensor data collected for the application, which for Mission Non-time Critical is generally
stored or for Mission Time Critical applications it is used as an input into the ‘bio-inspired’
formation control algorithm. Navigational data is determined from the collected localisation
data, which is required in both of the formation control algorithms that provide the trajectory
information for each vehicle. In addition, for USSN applications, irrespective of mission type,
vehicles will be operating in close range to each other and thus avoidance of vehicle collisions
is an important consideration.
Surface
Sea Floor
Swarm
Vehicles
Leader
24
Therefore, an understanding of the localisation techniques and the formation control algorithms
is required to determine the kind and quantity of data traffic expected in each of the application
classifications. Examples of payload sensor types and their mechanisms are also included.
2.6.1 Payload Data
There are a variety of sensors and mechanisms available to collect data for the different
applications and each of these have different data requirements. Table 2.2 provides a small
selection of underwater sensors that might be used for different applications. The amount of
payload data that is generated from these sensors can vary substantially.
For the purposes of this work, an initial assumption will be that the data for the Mission Non-
time Critical applications will be collected and stored and therefore will not impact on the
network traffic. For the Mission Time Critical applications the information gained from these
sensors will be included in the formation control algorithm, Rule 2, and can also be available to
be sent as lower priority data. In this case it will be aggregated to a manageable packet size not
more than 40 bytes plus overheads which will be further discussed in Chapter 5.
2.6.2 Localisation
Localisation data is required to determine position of a vehicle. Localisation can be classified
into two different approaches: absolute or relative positioning. Absolute positioning is where
vehicles need to know their actual geographical position, and this information is normally
acquired using GPS or similar positioning technologies. GPS can be problematic even in some
terrestrial settings, such as in forests or indoors, and also in large-scale deployments because
of power consumption and costs [12]. GPS positioning is practically impossible underwater, due
to the attenuation of radio waves.
Where absolute positioning is necessary during an underwater swarm operation, this
information can be obtained by:
• using one or more anchored nodes of known position. These anchor nodes often use a
broadcast approach, where the anchor nodes are located at certain intervals over the target
area and are used to inform other vehicles in the network of their known position; or
• For non real-time requirements, where vehicles are able surface and use GPS to capture
the vehicles absolute position. Generally, surfacing occurs at the beginning and end of a
mission but can also occur following a specific trigger. Various examples of using GPS for
absolute positioning has been implemented in underwater settings [22, 31, 98, 103, 146].
When vehicles only need to know the relative position of their neighbours, this can be gained
explicitly by having each vehicle send location details with their messages or implicitly by using
properties of the message itself. Implicit approaches include:
• Time of Arrival (TOA) technique which is based on measures of the travel time of a
message;
25
Table 2.2: Examples of Payload Sensor Types For Mission Time Critical and Non-time Critical Applications
Mission Time Critical
Application Area Sensor Type Mechanism
Pollution detection (e.g.
detection of crude oil leaks or
ballast discharge)
Active Acoustic Sensor
Hydrocarbon & Methane
Sensors
Fluorometer
• Active transmissions are reflected by boundaries
between different media. Larger droplets or plumes of a
leaking medium will give a stronger backscattered
acoustic signal.
• Dissolved CH4 molecules diffuse through a thin-film
composite membrane into the detector chamber, where
their volume is determined by means of IR absorption
spectrometry
• Detection and measurement of fluorescent compounds
such as Chloropyhll, CDOM, Crude Oil, and Fluorescein,
Rhodamine, and UV Tracer Dyes
Shipwreck or Black Box or
archaeological locator
Magnetometer • Measures disturbance in earth’s magnetic field
Mission Non-time Critical
Application Area Sensor Type Mechanism
Oil and Gas exploration – (e.g.
detection of hydrothermal
vents)
Methane sniffers • Two measurement principles exist for measuring
methane dissolved in water that are based on: dissolved
methane diffusing over a composite membrane into an
internal gas circuit where the CH 4 concentration is
measured with infrared spectrometry and directly into a
sensor chamber
Scientific/Environmental
sampling
Salinity, temperature,
depth or oxygen sensors,
phytoplankton
• Numerous instruments are used for measuring physical
aspects of the ocean for post mission analysis. E.g.
conductivity (salinity), temperature, pressure (depth),
dissolved oxygen, Chlorophyll A fluorometry
(phytoplankton) etc.
Surveying ocean bottom,
Bathemetry
Single beam echo
sounders
Multi beam echo
sounders,
Side scan sonars
Sub-bottom profilers
• The instruments currently used for this work require
large energy levels. This drives battery size and hull size
so that they cannot be currently integrated into small
swarming AUVs.
• New concepts are emerging where transmission of the
active sonar signal might occur from a “mother ship” and
the AUV swarm will be used to detect the return signal
from the seafloor.
• It can be anticipated that data will be required to be
distributed in the swarm (e.g. time and location) to
support post mission analysis.
26
• Time Difference of Arrival (TDOA) technique which is based on measures on the difference
of arrival time at different antennas;
• Angle of Arrival (AOA) technique that uses measurements of the relative angle between
nodes;
• Receiver Signal Strength Indicator (RSSI) technique that determines range from the power
in the received signal and the Doppler shift measurement of the relative inter-vehicle
velocity.
There are also some interesting biologically inspired alternative approaches suggested for
providing relative positioning underwater and some of these were discussed in Section 2.4.1.
2.6.2.1 Underwater Vehicle (AUV) Navigation
The most commonly used technique for navigation within an AUV underwater is a traditional
method of nautical navigation known as Dead-Reckoning [92]. To determine the position,
orientation and velocity of a vehicle, AUVs are generally equipped with an Inertial Navigation
System, which includes accelerometers, gyroscopes, doppler velocity technology (DVL) and a
magnetic compass. At the beginning of an operation the starting depth, latitude and longitude
are entered into the Inertial Navigational System that will perform the Dead Reckoning
calculations. The system then receives information from the on board navigational sensors that
measure motion along three or more axes enabling continual and accurate calculations of the
vehicle’s current depth, latitude and longitude. The advantages of this approach is that once the
starting position is set, the device does not need external information, which can be hampered
by poor weather conditions and, means for military operations, the vehicle cannot be detected
or jammed. The disadvantage of the dead reckoning approach is that errors in position will
accumulate because the current position is calculated solely from the previous position and that
vehicle drift, due to the impact of water currents or collisions for example are more difficult to
take into account. It can be assumed, however, that in most situations vehicles working in close
proximity will be subjected to similar motions and therefore their relative position will be
maintained. Generally this means that the geographical position of vehicles with inertial
navigation systems should be corrected from time to time with a local ‘fix' from other types of
navigation systems such as magnetic compass (for heading) or a GPS (latitude and longitude)
on the surface, particularly for long operations and non-time critical applications. Pressure
sensors are also included on-board vehicles depth measurements.
An example of a simpler and less computationally intensive approach, is the SeaVision™
prototype vehicle that uses only a LinkQuest NavQuest DVL and Compass [84, 86]. These
provide the localisation data that is feed into the main on-board computer in an 80 ASCII
character format and included: Pitch; Roll; Heading; Temperature; Velocity relative to current;
and velocity relative to bottom. The relative position, direction and velocity of the vehicle could
then be calculated. Aggregating the data is both practical and valuable to reduce the amount of
information sent around the swarm due to the high latency and low bandwidths of an
27
underwater acoustic channel. This data could be packaged into a message of no more than 40
bytes plus overheads discussed further in Chapter 5.
2.6.3 Mission Time Critical: Bio-inspired Formation Control Algorithms
There are numerous methods that have been developed to control a swarm of multi-vehicle
systems and they generally focus on predictive control either to maintain formations [49, 108,
110] or to help achieve a goal [71, 110]. Biologically inspired control or biomimicry has emerged
as a major field of research as it offers a way of accessing and emulating the intelligence,
principles of design, and the time-tested patterns and strategies observed in natural systems.
Artificial swarm research has developed out of the desire to provide autonomous control to
groups of robots and/or intelligent mobile devices and give control also to virtual swarms used in
computer programming for gaming and problem solving. Reynold's [110] Boid approach to
simulate flocking behaviour of birds is recognized as the beginning of the work in the field of bio-
mimicry to replicate natural swarming into artificial swarms and is the forerunner to the field of
Particle Swarm Optimisation (PSO) [75] and Ant Colony Optimisation (ACO) [35, 38].
Dorigo [37] developed a method of finding an optimal path in a graph, based on the behaviour
of ants searching for a food source. It was found that they lay down a track of pheromone on
their return from the food source. The shortest path means the quickest return time and
therefore the chance that more ants will have returned on that path and therefore the
pheromone level will be higher. An adaptation of this idea was used by the SHOAL project [122]
as discussed in Section 2.4.1. This original idea has since found an increasing number of
applications, including scheduling problems, the Travelling Salesman Problem, and the vehicle
routing problem found in packet-switching connectionless networks like the Internet. One
application of PSO in communications has been on minimising energy consumption by
optimising clustering of nodes in ad-hoc networks, however it is recognised that it introduces
enormous traffic and computational overheads to a network [51].
The Boids [109] approach to simulating flocking of birds has inspired many of the more recent
attempts of replication of bio-inspired behaviour into artificial swarms in underwater settings.
The Boid model is based on a set of three rules that individual agents use to evaluate and script
their path within the swarm:
• Rule I: Collision Avoidance: steer for separation to avoid nearby agents
• Rule II: Velocity Matching: steer for alignment and to match nearby agents’ velocities
• Rule III: Flock Centering: steer to stay close to nearby agents
These three simple rules have inspired much of the work in swarm flocking behaviour including:
• Wilensky Flocking model or NetLogo [140] which is an open source online swarming
model for animated motion of flocks;
• Vicsek model [136] that introduces the term Self-Propelled Particles (SPP); and
28
• Various implementations of the attraction-repulsion model (ARM) that focus on Rules I
& III.
In Chen’s [22] work the difference in range between vehicles is used as a ‘force’ in a more
traditional Attraction/Repulsion model (ARM), which is a variation of the Boids flocking model.
Chen creates the concept of a single ‘virtual potential energy' variable similar to a spring system
in physics to represent the ‘attraction’ and ‘repulsion’ forces used to calculate each vehicle’s
new trajectory. The proposed team formation uses a hybrid approach depending on whether the
absolute or relative positions of vehicles are known.
Where the actual position information of other vehicles in the neighbourhood is known, an
Absolute Formation Adjustment algorithm is proposed and this replicates Rules I & III of the
Boids model. If relative speed information is also obtained, which, it is suggested, can be gained
indirectly from estimating the Doppler frequency shift in the message arrivals as discussed
above in RSSI, then a Relative Formation Adjustment algorithm is used, which incorporates
Rule II.
The virtual potential energy concept is that a vehicle can calculate its pitch and yaw using the
positions of the other vehicles in the swarm, which is obtained from regular communication
exchange This is calculated by solving a minimisation of a virtual potential energy metric Eij,
where i is the vehicle and j = 1,2,…N with N vehicles in the swarm. That is, 𝐸𝑖𝑗 = 12 ∑ 𝑘𝑖𝑗 𝑗∈𝑁𝑖 ∆𝑥𝑖𝑗2
where kij is the virtual spring constant and ∆𝑥𝑖𝑗2 is the displacement from the expected formation
equilibrium between i and j [22]. A similar minimisation is done for the Relative Formation
Adjustment algorithm except that an additional relative velocity term is included.
Other researchers have investigated variations of the Boids model by using Attraction/Repulsion
models (ARM) [72, 139, 145] while Othman [97] used the Wilensky Flocking Model and the
NetLogo simulation tool. Yang proposed dividing up the ‘maximum sensing range’ of a vehicle
(a circle around the centre of the vehicle) into 5 regions, two to the front of the vehicle, one to
either side of the vehicle and one to the rear. The vehicle of interest then chooses 5
neighbouring vehicles, one from each region, to calculate its new trajectory by using a Motor
Schema [7] school vector, Vschool:
𝑉𝑠𝑐ℎ𝑜𝑜𝑙 = 𝑤𝑓𝑟𝑜𝑛𝑡�𝑣𝑓𝑙𝑒𝑓𝑡 + 𝑣𝑓𝑟𝑖𝑔ℎ𝑡� + 𝑤𝑙𝑎𝑡𝑒𝑟𝑎𝑙�𝑣𝑙𝑙𝑒𝑓𝑡 + 𝑣𝑙𝑟𝑖𝑔ℎ𝑡� + 𝑤𝑟𝑒𝑎𝑟𝑣𝑟𝑒𝑎𝑟
Where wfront, wlateral, and wrear are weightings for each area and vfleft, vfright vlleft vlleft and vrear are the
vector directions of each of the five chosen neighbourhood vehicles, which need to be
communicated back to the centre vehicle. Yang’s work did not cover, nor did he report on, how
this system could be expanded for the 3D operations that are experienced underwater.
These models show that even without any centralised control, the direction of all agents
converged eventually. The algorithms proposed for Boids, ACO & PSO assume that there are a
very large number of agents in the network, as is defined in Criteria 1 in Section 2.3.1 a
definition of a swarm. Yet for underwater swarms we conjecture at this time that smaller number
of vehicles will define an underwater swarm due to economic and communication limitations.
29
This is because AUV's and gliders, although reducing in cost are still very expensive and also
due to the difficult communication channel characteristics particularly around low bandwidths
and slow signal speeds.
The other assumption that is made with these biologically inspired formation control algorithms
is that the agents’ positions are known or can be established easily. Agent localisation with
respect to other agents in the swarm plays an important part of any formation control algorithm
as has been discussed in Section 2.6.1. Therefore, an important research question, is: What is
the maximum number of vehicles that a swarm can support in a single-hop network, as defined
on the x-axis of the taxonomy in Figure 2.1, based on the ability of the vehicles to determine
and exchange localisation information among swarm members.
For the purposes of this work, the data that each vehicle needs to distribute is it position and
planned trajectory (heading). Once each vehicle receives neighbouring vehicles information it
can calculate its new trajectory based on the formation control algorithm that is implemented.
2.6.4 Mission Non-Time Critical: Patterned Formation Control Algorithms
For the Mission Time Critical applications that study dynamics oceanographic phenomena
where measurements are required to be taken either at different sites at the same time or same
site at different times, a structured pattern formation control of a multi-vehicle system is best
suited. There have been several researchers that have investigated algorithms for patterned
cooperative mapping and sampling [22, 86, 98, 123, 146]. Classic formations are lawn [86] or
lane [146] patterns, which provide a structured sweep of an area. Mare showed that provided an
AUV obtains a regular estimate of its current position (using an update every 200ms) and has a
target location to aim for that it can self adjust to follow a line. In a multi-vehicle swarm,
communication between vehicles can support improved navigation as it allows techniques that
employ data from more than one source, which provides more accurate positioning. In addition,
the ability for vehicles to share information has been recognised to allow a reduction in the need
for expensive resources (such as gyro compasses and high-resolution bathymetry sonars) in
every vehicle [77, 123].
2.7 USSN Communication Challenges The challenges of designing an explicit communication system for an Underwater Swarm
Sensor Network (USSN) have been discussed in this chapter. The ultimate purpose of the
explicit communication exchange between vehicles within a swarm is to maintain swarm
synchronisation amongst the vehicles and complete the mission task. Swarm synchronisation
refers to the ability to maintain a successful swarm-like formation of the vehicles and means that
vehicles can successfully manoeuvre within close proximity to each other without vehicle
collisions and be able to, using decentralised control, know the overall direction of the swarm
and the current mission status.
This closing section will examine the major research questions that will be investigated in this
work, summarising the Swarm Network Characteristics and Data Traffic Requirements.
30
2.7.1 Swarm Network Characteristics
An analysis of the information presented in this chapter led to the identification of the design
criteria that must be met when developing an explicit communication protocol for a closely
operating single-hop Underwater Swarm Sensor Network. The criteria are:
• A group of computationally simple vehicles
• Decentralised network with no master controller
• Scalable to deal with device failures, loss or re-joining the swarm
• Self-configuring to enable ease of deployment of an autonomous network
• Self-correcting in adapting to the harsh channel conditions
• Point-to-multipoint or ‘everyone-to-everyone’ messaging exchange
• Swarm synchronisation requiring continuous communication exchange
• Latency-aware, for quick time-sensitive information distribution
• Cognitive of the application-specific requirements and mission objectives
2.7.2 Summary of USSN Data Traffic Requirements
In the majority of wireless ad-hoc sensor networks and for mobile underwater DTNs, the
communication traffic is generally sporadic and usually point-to-point. This is very different to
the communication traffic expected in a single cluster swarm of autonomous underwater
vehicles that require:
• continuous traffic with quick local and global information dissemination;
• time-sensitive information distribution;
• maintenance of a self-configuring network while collecting mission data; and
• full network connectivity, with successful communication mode ‘everyone-to-everyone'.
Table 2.3 Single Cluster Underwater Swarm Sensor Network (USSN) Traffic Characteristics
Application
Category Mission time critical Mission non-time critical
Traffic
Characteristics Critical time-sensitive
information from multiple
vehicles compete for the
same network resources
Variable Link
Quality
Time-sensitive information
from multiple vehicles
compete for the same
network resources
Variable Link
Quality
Communication
Requirements Fairness across all
vehicles, high connectivity
Guaranteed data
delivery
Good connectivity Reliable link layer
transmissions
31
The main traffic characteristics and communication requirements and comparison between the
two application classifications are summarised in Table 2.3. This is developed further in Chapter
5 with MAC layer Quality of Service requirements.
2.7.3 Research Goal and Specific Research Questions
In order to address the research questions discussed in Chapter 1, further research questions
arise related specifically to the communication protocol. The overall goal is to design and
develop an effective communication protocol for an underwater swarm of AUVs and in
particular, to determine what the limitations are on the size and coverage area of an operating
swarm based on underwater conditions.
Thus the more specific research questions are:
• What are the limits to the number of vehicles in a densely operating underwater swarm,
based on various coverage areas, given the ability to explicitly exchange inter-vehicle
data through an acoustic communication network?
• How can a Medium Access Control (MAC) communication protocol be designed to take
into account the constraints of a short-range underwater acoustic channel?
• What are the properties and limitations of the MAC protocol and how do they impact on
swarm synchronization? And the operating size of a swarm?
o What is an acceptable level of delay for full exchange of information between
vehicles and how does that impact on the QoS requirment?
o What throughput levels are required to maintain network and node connectivity,
which are required for swarm synchronisation?
o How does packet loss impact on swarm synchronization? and What is a
suitable design approach to minimize the effects of packet loss to maintain
swarm synchronisation?
o How can the MAC protocol deal with packet size and range?
2.8 Conclusion There has been substantial research and even development into swarm networking and its
communication requirements for terrestrial use, yet the research and development of
underwater swarm operations is still in its infancy. With only a few projects that have begun to
address the many engineering challenges of underwater swarm networks there is significant
development work still to be done in the area of communication requirements and protocols for
underwater swarms. This chapter has provided an overview of these studies and projects while
focusing on the challenging issues for the development of a communication protocol.
Swarm robotic systems are considered to be: cooperative (common task); aware (of other
robots); coordinated (taking into account actions executed by others); and decentralised (lack
of central control agent). The study and adoption of biologically inspired techniques, which
access nature's intelligence and principles of design, can provide valuable insights into the
developments of the networks and the communication required between vehicles. Critical to
32
robotic swarms, where vehicles are operating in close range with each other, is some form of
communication between vehicles.
The swarm formation control algorithms, whether they are biologically inspired or more
structured patterned arrangements, require some form of spatial location information from
neighbouring vehicles to function. This requirement provides the development of the traffic
characteristics and communication requirements for a USSN that have been briefly discussed in
this chapter and will be expanded upon in Chapter 5 with the performance metrics.
Because of the slow propagation speed of sound, the very low bandwidths, and the fading
characteristics of the communication channel, the range between vehicles will have a large
impact on the quality of the communication between them. The range will also have an impact
on the importance of the timing of the communications required to avoid the potential for
vehicles to collide into each other. That is, with sparse deployment, the condition related to
vehicle collision is reduced or negated and therefore more sporadic communication is
acceptable.
Chapter 3 will present the study of the characteristics of the short-range acoustic channel as
there were no short-range underwater acoustic channel models available when this research
begun. This work is the foundation for the design and development of the new protocols as well
as the analysis and evaluation of the performance of the protocols in non-ideal conditions.
33
Chapter 3 Short-Range Underwater Acoustic Communication Channel
Characteristics
3.1 Introduction An understanding of the short-range underwater acoustic communication channel and its
properties is necessary for the design and development of MAC (Medium Access Control)
protocols to support communication amongst a group of swarming Autonomous Underwater
Vehicles (AUV). As discussed in the last chapter, it is important to optimise the use of the limited
resources available at all levels of the protocol stack and thus clearly identifying the channel
constraints allows strategic trade-offs and best use of these limited resources. In addition,
understanding the communication channel is necessary for developing the non-idealistic model
in which to simulate and test the new protocol designs.
There has been little work done on the short-range acoustic channel model, as there has not
been the operational demand for these systems. Recently the developments in underwater
acoustic sensor networking (UW-ASN) and the use of multi-hop networking architecture and
data muling operations have generated interest within the research community to develop
shorter range underwater communication systems [4, 41]. The progress towards low cost AUVs
will continue this interest with the additional need to account for mobility of AUV vehicles. As the
knowledge of long-range channel models is well established, the characterising of a short-range
channel model will initially extrapolate this understanding.
This chapter is divided into two major components, the analysis of an underwater acoustic
channel for communication purposes and then the modelling of the channel characteristics in
which the communication system will operate within. The first part of the chapter will examine
and analyse the characteristics of a short-range underwater acoustic channel based on the
long-range physical models that are already well established. The second half of the chapter will
determine the transmission channel state information and appraise the physical layer
characteristics that will be required to test and evaluate the communication protocols in a
simulation environment.
3.2 Acoustic Channel The underwater acoustic communication channel is recognised as one of the harshest
environments for data communication, with long-range calculations of optimal channel capacity
of less than 50kbps for SNR of 20dB [125]. Predictability of the channel is also very difficult with
the conditions constantly changing due to seasons, weather, and the physical surroundings of
sea floor, depth, salinity and temperature.
34
Figure 3.1: Underwater Acoustic Environment
A typical acoustic underwater environment for point-to-point data transmission using a single
projector (transmitter) - hydrophone (receiver) pair is illustrated in Figure 3.1. The acoustic
channel characteristics for data communication are quite different to those of RF that is
generally used in a terrestrial setting due to the properties of both the medium itself – seawater
– which is approximately 1000 times denser than air and the use of sound, a pressure wave.
These facts together with the unique noise environment, which is frequency dependent;
attenuation, which is both range and frequency dependent; and the different multipath
characteristics due to channel geometry will be explored in more detail below.
A simple schematic of the Projector and Hydrophone functional block diagrams is shown in
Figure 3.2. The Projector accepts Packets to be transmitted at the Data Source and these are
carrier frequency is modulated in sympathy with the data bits. The modulated signal is amplified
to a level sufficient for signal reception at the hydrophone. The acoustic power radiated from the
Projector is a fraction of the electrical power supplied to it, and this is represented by the
electrical to acoustic conversion block in the Projector and by the acoustic to electrical
conversion in the Hydrophone. The efficiency or sensitivity of underwater transducers (ηtx or ηrx )
quoted by manufactures and researchers vary significantly from as low as 20 % to 70 % [137].
This work will use 50% as a typical value. On the receiver side, general carrier recovery will be
used for signal detection and amplification to occur on discernible signals that is based on
overcoming a SNR (mean signal power to mean noise power ratio) detection threshold set by
modulation and bit error correction. The received data packets will then be available for use
35
Figure 3.2: Block Diagram of a Projector and Hydrophone
within the higher layers of the OSI communication stack for input into the vehicles control and
navigation requirements and/or data storage.
3.3 Underwater Acoustic Channel Characteristics This section will investigate the parameters and characteristics of an ocean channel that will
affect the acoustic signal propagation from a Projector to a Hydrophone. There are well-
established long-range underwater acoustic channel models that will be discussed to derive and
present the data transmission characteristics for a short-range link that will be used in this work.
3.3.1 Acoustic signal level
The projector source level, SPLprojector, is generally defined in terms of the sound pressure level
at a reference distance of 1 m from its acoustic centre. The source intensity at this reference
range using a total acoustic power consumption of Ptx, is:
𝐼 = 𝑃𝑡𝑥𝐴𝑟𝑒𝑎
Wm-2 (3.1)
and measured in dB 're 1 μPa' but strictly meaning 're the intensity due to a pressure of 1 μPa'.
The standard reference wave intensity used for underwater sound is:
𝐼𝑟𝑒𝑓 =𝑃𝑎𝑟𝑒𝑓2
𝜌∗𝑐 Wm−2 (3.2)
where the reference pressure level Paref is 1 μPa, ρ is the density of the medium and; c is the
speed of sound (averages for sea water are: ρ = 1025 kg/m3 and c=1500 m/s) (Coates, 1989;
Urick, 1967).
36
3.3.2 Transmitter (Projector) Signal Intensity
Thus the transmitter acoustic signal level, SPLprojector, at 1 m from an omni-directional projector
(surface area (sphere) measured at 1 m of 4πr2 = 12.6 m2) can be written:
𝑆𝑃𝐿𝑝𝑟𝑜𝑗𝑒𝑐𝑡𝑜𝑟 = 10𝑙𝑜𝑔 �(𝑃𝑡𝑥 12.6⁄ )𝐼𝑟𝑒𝑓
� + 10𝑙𝑜𝑔(𝜂𝑡𝑥) + 𝐷𝐼𝑡𝑥 dB re 1 μPa (3.3)
where ηtx represents the efficiency of the Projector and 𝐷𝐼𝑡𝑥 the Projector directivity index which
is determined by:
𝐷𝐼𝑡𝑥 = 10𝑙𝑜𝑔 � 𝐼𝑑𝑖𝑟𝐼𝑜𝑚𝑛𝑖
� (3.4)
where Iomni is the intensity if spread spherically and Idir is the intensity along the axis of the
beam pattern. Directivity of a Projector can concentrate the transmitted source sound level in a
given direction, which is similar to the concept of antenna gain in RF. Thus;
𝑆𝑃𝐿𝑝𝑟𝑜𝑗𝑒𝑐𝑡𝑜𝑟 (𝑃𝑡𝑥, 𝜂𝑡𝑥,𝐷𝐼𝑡𝑥 ) = 170.8 + 10𝑙𝑜𝑔(𝑃𝑡𝑥) + 10𝑙𝑜𝑔(𝜂𝑡𝑥) + 𝐷𝐼𝑡𝑥 dB re 1 μP (3.5)
3.3.3 Signal Attenuation
Sound propagation in the ocean is influenced by the physical and chemical properties of
seawater. Signal attenuation or propagation loss is the measure of the lost signal intensity from
Projector to Hydrophone and includes a spreading and absorption component. Accurate
modelling of propagation loss is difficult due to its dependency on the properties of seawater
that are constantly changing from day to day and location to location. In addition, parameter
variations that may occur in short range operation will be particularly focused on.
3.3.3.1 Spreading loss
Spreading loss is due to the expanding area that the sound signal encompasses as it
geometrically spreads outward from the source.
𝑃𝐿𝑠𝑝𝑟𝑒𝑎𝑑𝑖𝑛𝑔 (𝑟) = k 10𝑙𝑜𝑔(𝑟) dB (3.6)
where r is the range in meters and k is the spreading factor.
When the medium in which signal transmission occurs is unbounded, the spreading is spherical
and the spreading factor k=2, whereas in bounded spreading, considered as cylindrical k=1.
Urick [134] suggested that spherical spreading was a rare occurrence in the ocean but
recognised it may exist at short ranges.
37
3.3.3.2 Absorption loss
The absorption loss is a representation of the energy loss in the form of heat due to the viscous
friction and ionic relaxation that occurs as the wave generated by an acoustic signal propagates
outwards and this loss varies linearly with range as follows:
𝑃𝐿𝑎𝑏𝑠𝑜𝑟𝑝𝑡𝑖𝑜𝑛 (𝑟, 𝑓) = 10𝑙𝑜𝑔(𝛼(𝑓) × 𝑟 × 10−3) dB (3.7)
where r is range in metres and α(f) is the absorption coefficient.
There are three dominant effects that cause the absorption of sound in seawater: viscosity
(shear and volume), ionic relaxation of boric acid and magnesium sulphate (MgSO4) molecules
and the relaxation time. The effect of viscosity is significant at high frequencies above 100 kHz,
whereas the ionic relaxation effects of magnesium affect the mid frequency range from 10 kHz
up to 100 kHz and boric acid at low frequencies up to a few kHz. In general, the absorption
coefficient, α(f), increases with increasing frequency and decreases as depth increases [37,
118] and is significantly higher in the sea compared with fresh water due predominately to the
ionic relaxation factor. Extensive measurements of absorption losses over the last half-century
have lead to several empirical formulae which take into account frequency, salinity,
temperature, pH, depth and speed of sound.
A popular version is Thorp's expression [132], Equation 3.8, which is based on his initial
investigations in the 1960's and has since been converted into metric units (shown here). It is
valid for frequencies from 100Hz to 1MHz (f is in kHz) and is based on seawater with salinity of
35% ppt, pH of 8, temp of 4◦C and depth of 0 m (atmospheric pressure), which is assumed but
not stated by Thorp.
𝛼(𝑓) = 0.11𝑓2
1+𝑓2 + 44𝑓2
4100+𝑓2+ 2.75 × 10−4𝑓2 + 0.0033 dB/km (3.8)
A more recently proposed variation of the absorption coefficient include Fisher and Simmons
[47] who in the late 1970's found the effect associated with the relaxation of boric acid on
absorption and provided a more detailed form of absorption coefficient α(f,d,t) in dB/km which
varies with frequency (f), pressure (based on depth in meters, d) and temperature (t) in ◦C (also
valid for 100 Hz to 1 MHz with salinity 35% ppt and acidity 8 pH) [47, 118], given in Equation
3.9.
𝛼(𝑓,𝑑, 𝑡) = 𝐴1𝑓1𝑓2
𝑓12+𝑓2 + 𝐴2𝑃2𝑓2𝑓
2
𝑓22+𝑓2+ 𝐴3𝑃3𝑓2 dB/km (3.9)
The 'A' coefficients represent the effects of temperature, while the 'P' coefficients represent
ocean depth (pressure) and f1, f2 represent the relaxation frequencies of Boric acid and
(MgSO4) molecules. These terms were developed by Fisher and Simmons [47] and presented
more recently by [37, 118] and can be found in Appendix A.
38
Figure 3.3: Absorption Coefficient vs Frequency
Figure 3.3 shows the absorption coefficients in dB/km vs signal frequency for both Thorp and
Fisher & Simmons coefficients and shows that in general Absorption Coefficient increases with
increasing frequency at any fixed temperature and depth. Up until around 80kHz, temperature
change has a more significant affect on the Absorption Coefficient than depth, but above these
frequencies depth begins to dominate [37,118]. In any case, Thorp’s 'approximation' is quite
close to Fisher and Simmons and is more conservative at the frequencies shown as the
Absorption Coefficient can be almost 10dB higher. Sehgal [118] shows that at higher
frequencies above 300kHz, Thorps model predicts lower losses, as it does not take into account
the relaxation frequencies found by Fisher and Simmons. If depth and frequency are fixed and
temperature varied from 0 to 27 ◦C, there is a decrease in the Absorption Coefficient of
approximately 4 dB/km for frequencies in the range of 30 to 60kHz, which correlates to work
presented by Urick [134] (Fig5.3 pg 89).
If we consider where AUV swarms are most likely to operate, in the 'mixed surface layer', where
temperature varies considerably depending on latitude [69], temperature may be an important
factor. It should be noted that if operating in lower temperatures the Absorption Coefficient is
higher and thus using 0°C will be a conservative alternative. At shorter ranges, the significance
of the Absorption Coefficient is expected to be less due to the linear relationship with range from
Equation 3.7, which will be discussed further in the next section.
As mentioned, depth (or pressure) has less of an effect on the Absorption Coefficient than
temperature at these lower frequencies. Domingo [37] investigates the effect of depth on
absorption and confirms that for lower frequencies of less than 100kHz there is less change in
39
the Absorption Coefficient. More specifically Urick [134] defined the variation by α(d) but has
also suggested an approximation of a 2% decrease for every 300 m depth.
𝛼(𝑑) = 𝛼 × 10−3(1 − 5.9 × 10−6) 𝑑 dB/m (3.10)
where d is depth in meters.
Thus, depth variations are not expected to play a significant role in short range AUV swarm
operations.
3.3.3.3 Propagation loss
Total propagation loss is the combined contribution of both the spreading and absorption
losses. Urick [134] established that this formula of spreading plus absorption yields a
reasonable agreement for long-range observations.
𝑃𝐿𝑙𝑜𝑠𝑠(𝑟, 𝑓,𝑑, 𝑡) = k 10𝑙𝑜𝑔(𝑟) + 𝛼(𝑓,𝑑, 𝑡) 𝑟 × 10−3 (3.11)
where r and d in meters, f in kHz and t in ◦C.
For short-range operation, longer range models have been used in this work and adapted for
the shorter ranges defined in this work as there are no short-range models developed. This is
justified due to the short ranges being discussed are much greater than the wavelength at
40kHz (37.5mm) and therefore based on this assumption the pathloss component will follow a
similar trend to long range predictions.
(a) Signal Attenuation showing spherical spreading and
absorption factors
(b) Comparing Absorption Models using spherical
spreading. Frequency change shown using Thorp
Model and Temperature °C and Depth m changes
shown in Fisher and Simmons Model
Figure 3.4: Path Loss vs Range
40
Figure 3.4(a), illustrates the contribution of the spreading and absorption terms for increasing
range and also the variation of the absorption coefficient with signal frequency. It can be
observed that for very short communication ranges of less than 500 m, the absorption term is
less significant than the spreading term even at higher signal frequencies and therefore the
spreading factor k has a more significant impact on the total Propagation Loss at these ranges,
with signal attenuation lower for cylindrical spreading than spherical spreading. At short range
and with onmi-directional antennas, spherical spreading will be assumed in USSN operations.
In Figure 3.4(b) the total Path Loss is shown at shorter ranges of up to 500 m and the Thorp
model gives a more conservative value over this distance for both signal frequencies (40 kHz
and 60 kHz) used. Also illustrated are changes in depth and temperature in the Fisher and
Simmons model for a signal frequency of 40 kHz however provides some insight into the
variations due to depth and temperature.
In summary using the two models, Thorp and Fisher & Simmons, the two important
characteristics that can be drawn from propagation loss at the short ranges of interest for AUV
swarm operation are:
• spreading loss dominates over absorption loss in the total propagation loss, and thus
the 'k' term has a significant impact on the attenuation of the signal at very short ranges
(less than 50 m) as illustrated in Figure 3.4 (a). While the range between vehicles in an
AUV swarm operation is much less than depth, spherical spreading can be assumed;
and
• for short range (up to 500 m) the frequency component of absorption loss is the most
significant compared with the possible temperature and pressure (depth) changes as
seen in Figure 3.4(b). In addition, as range increases the value of absorption loss also
increases at an increasing rate, effectively meaning that the communication channel is
band-limited and available bandwidth is a decreasing function of range. Thus, choosing
an appropriate signal frequency for a range is very important.
3.3.3.4 Speed of Sound
The speed of sound underwater and its profile impacts on its use as a communication medium.
It’s slow speed (compared with RF) at 1500 m/s typically is particularly important in designing a
communication protocol as will be discussed further in the following chapters. In addition, as a
result of the unique sound speed structure in the oceans which create temperature gradient
channels that trap acoustic signals a phenomena call ray bending occurs.
A propagating acoustic signal in an underwater channel bends according to Snell's Law, to
lower signal speed zones. Figure 3.5 shows a typical ocean Sound Speed Profile, although
variations occur with location and seasons. The profile is depth dependent, where sound speed
is influenced more by temperature in the surface layers and by pressure at greater depths. Ray
bending is thus more accurately defined by the location of the transmitter and receiver within the
41
Figure 3.5: Typical Sound Speed Profile in the Ocean [29]
geometry of the channel and in particular their depths. For short-range operations the variation
in the speed of sound will be minimal. Thus in this work, a constant speed of sound of 1500 m/s
will be used in all calculations.
3.3.4 Underwater Multipath Characteristics
Sound propagation in the ocean is influenced by the geometry of the channel. Depending on
this channel geometry, sound signals will experience various reverberation mechanisms that will
result in multiple signal paths. When these multiple signals are being received at a hydrophone,
they will have different arrival times with different propagation losses.
Multipath signals, in general, represent acoustic energy loss, however, for communication
systems it is the Inter Symbol Interference (ISI) that can occur which is particularly detrimental
as it can significantly increase the error rate in a receiving signal. It is this delay spread of the
signal component arrivals that can cause ISI to occur if they overlap with previous or future
signal arrivals that will cause symbol corruption or loss and therefore bit errors. As the speed of
sound propagation is very slow in an acoustic channel, this delay spread can be significant.
The hydrophone (receiver) may receive the direct signal and a combination of various multipath
signals that have been reflected, scattered or bent. It is these multiple components of the signal
that are delayed in time due to the various path lengths that may create ISI and errors in symbol
detection.
Multipath signal formation is dependent on the location of the projector (transmitter) and
hydrophone within the geometry of the channel. There are several known mechanisms
responsible for creating multipath signals underwater for single transmitter-receiver pairs. These
42
are a result of reverberation, which refers to the reflections and scattering of the sound signal
[27, 37, 44, 134]:
• Multipath propagation caused by boundary reflections at the sea-floor or sea-surface,
seen in Figure 3.1.
• Multipath propagation caused by reflection from objects suspended in the water, marine
animals or plants or bubbles in the path of the transmitted signal
• Surface scattering caused by sea-surface (waves) or sea-floor roughness or surface
absorption, particularly on the sea bottom depending on material
• Volume scattering caused by refraction off objects suspended in the signal path.
Reverberation and the resultant multipath signals can be particular severe in shallow water and
will occur in deep water if the transmitter and receiver are located near the surface or bottom.
Thus the geometry of the channel being used is a major determinate of the number of
significant propagation paths and their relative strengths and delays.
For very short-range channels that will be used in AUV swarm operations, multipath will be
affected by the range-depth ratio, which is expected to produce fewer multipath signals at the
hydrophone from surface and seafloor reflections [55, 100, 136]. It has been found that some
improvement can be gained through directing the beam of the transmitted signal and the
directional properties of the receiver [137], however this will require an additional level of
complexity for mobile AUV's due to the need for vehicle positioning before sending or when
receiving a signal.
The application of swarms of AUV’s where multiple vehicles are operating at very close range
and requiring continuous communication amongst the vehicles using a broadcast approach will
mean that there will be many vehicle surfaces within the network that can cause sound echo’s
and scattering. It is proposed that this sound reflections off vehicles within the swarm can
become significant enough to deserve a new reverberation category; swarm reverberation.
3.3.4.1 Swarm Reverberation
The application of active sonar has motivated a significant amount of work in understanding
reverberation underwater. Active sonar uses a pulse (ping) or sequence of pulses to detect,
localise and classify underwater contacts based on a return signal (reflection) from the contact.
The background against which the returned signal needs to be detected contains reverberation
in addition to the ambient noise and self-noise at the receiver.
The scenario for USSN’s is different from active sonar in that instead of identifying a single or
short sequence of pulses, for the USSN scenario the communication hydrophones need to
receive a relative long series of pulses that make up a full packet of bits. These are received on
a frequent basis from omni-directional antennas, with the communication protocol endeavouring
to make full utilisation of the channel. Having so many vehicles at short range will create many
reflective boundaries in a similar manner to volume scattering where the size of the scattering
43
Figure 3.6: Data Transmission and Swarm Reverberation from a 4 vehicle USSN
Table 3.1: Packet Timing Diagram with Swarm Reverberation Reflections for a 4-vehicle Swarm at 30m for Different Packet Sizes
44
particles, d (a vehicle), will be larger than the wavelength (λ). When d >> λ the scattering is
independent of frequency and depends only on the acoustic properties of the scattering and its
cross-section or geometrical acoustic scattering [44 pg291].
Figure 3.6 illustrates a 4 vehicle USSN where Vehicle 1 (V1) transmits a data packet with an
omni-directional antenna that will be received by vehicles V2, V3 and V4 at various delays
depending on the range as shown in Table 3.1. V1’s data packet arrives at V2 in 20 ms and if it
is a 72 byte packet it will complete reception at 80 ms (at a data rate of 9600 bps). V2 is given
permission to access the channel next and it sends outs its data packet immediately upon
completion of reception of V1’s packet. Each vehicle has its turn to access the medium.
Due to the size of each vehicle, there will be sound energy reflected off the vehicles surfaces.
V2, V3 and V4’s scattered energy will potentially return to V1 as is desired in active sonar
applications but will also scatter in the direction of other vehicles. Figure 3.6 illustrates the
reverberation echo’s from V3 back to V1 and to V2 and V4. Only the first reflection is shown and
included in this analysis. The energy from the reverberation off V3 will begin arrival at V2 at 48
ms well before the completion of reception of the data packet from V1. Table 3.1 illustrates the
timing of reverberation for two packet sizes of 72 and 36 bytes. It can be seen that the
relationship between range (propagation time) and packet length (transmission time) will impact
on the amount of overlap between the data signal arriving at a vehicle and the swarm
reverberation experienced at that vehicle during reception. The swarm reverberation signal
strength will need to be incorporated as background interference from which the data signals
will need to be detected.
Scattering and reflected signal strength will depend on the surface material used on the vehicles
and the incident power. In the analysis, we have assumed perfect reflection and anti-phase
signals so no signal cancellation will occur.
3.3.4.2 Reverberation
Reverberation is described as the persistence of sound following the arrival of the intended
sound front at a receiver and is a long, slowly decaying quivering tonal noise that is particularly
obnoxious in systems of high power and or low directivity [134]. High power will not be an issue
in swarm vehicles due to the low transmitter powers required, however the requirement of omni-
directional operation will mean that reverberation is of some concern. Any of the various
multipath mechanisms can create complex scattering of sound and will cause interference for a
transducer. For example, the seafloor, like the sea surface is an effective reflector of sound and
the roughness of the sea surface with the trapping of air bubbles just under the surface will
make it a complex scatterer of sound in both time and space.
Various empirical models are available to represent reverberation depending on whether it is
due to sea surface, under ice, sea floor or volume scattering [44, 134, 137]. Lambert’ Law
represents the relationship between scattering strength and grazing angle for sea-floor
reverberation [44, 134] as Chapman and Harris for sea-surface [44, 137]. Each type of
45
reverberation in addition to grazing angle has different dependencies such as the ‘classification
of the sedimentary composition’ of the sea-floor, the ‘wind speed’ that provides a measure of
the roughness of the sea-surface and ‘depth’ for volume scattering. Volume scattering is not
uniformly distributed in the sea and scattering strength vs depth profile changes with frequency,
location and time and tends to concentrate at deeper layers below 500 m where scattering
strengths are estimated at over -80 dB. In the mid-layer ocean it is expected to be in the range
of -50 to -70 dB [137]. For sea surface scattering Chapman and Harris and later Chapman and
Scott [44] showed that the experimental measurements fit quite well their empirical formula,
particularly between frequencies of 1 to 10 kHz and at higher frequencies there was less
variation. The recommended scattering strength for use in sonar design is around the -40dB for
high sea state and -50 dB for low sea states [44, 107, 137].
The analysis done on reverberation strength in underwater systems have been focused on
active sonar and the detection of a target from the expected strength of the reverberation of the
source signal used. The scenario in this work is different in that it is the reverberation strength at
the receiver of a transducer that is receiving a full packet of bits in a period of time (such as 424
bits using 9.6 kbps is 44ms). The transducer in this case is receiving a series of sound pulses
that each have a decaying energy level due to reverberation that the signal experiences on the
way to the receiver and as it passes the receiver. This is a complex field, which is made worse
by the fact that it is greatly affected by the environment and location of the transmitter and
receiver.
Thus it is proposed that this will impact significantly on the ability of a receiver to detect a signal
in this interference and thus SNIR (Signal-to-Noise+Interference-Ratio) will be used instead of
SNR (Signal-to-Noise-Ratio). An average Reverberation level (RL) of -46 dB for high sea states
and -54 dB for low sea states [137 pg113], based on the backscattering strength, will be used
for the interference value with a variability of ± 3 dB.
3.3.4.3 Reverberation and Transmitter Power Levels
A major challenge related to reverberation level, especially for data communication, is its
dependency on the projector transmitter power, Equation 3.23, so that the more sound energy
transmitted into the water the more sound energy will be reflected back to the receiver. Thus, it
is transmitter power changes that will be of particular interested.
3.3.4.4 Delay Spread and Coherence Times
In an acoustic underwater channel, reverberation causes both time-invariant and time varying
multipath. Time-invariant deterministic propagation multi-path models have been developed for
the various reflective and ray bending path options. These are significant in themselves with
multipath spreads in the order of 10 to 100 ms [125], which can have a significant impact on the
ISI. Take for example Figure 3.1 with the projector and hydrophone separated by 50m and both
at a depth of 50m, the delay spread between the direct path and the first sea surface reflection
is ≈ 41.2 ms (packet arrivals are at ≈ 33.3 ms and 74.5 ms respectively). Assuming a packet is
46
424 bits and using a data rate of 9.6 kbps gives a transmission time of 44.2 ms which means
that the direct path packet will still be completing arrival (at 77.5 ms) when the first reflection
arrives at 74.5 ms.
Multipath in an underwater channel also will be due to time-varying components caused by the
surface or volume scattering or by internal waves in deep water that are responsible for random
signal fluctuations. There has been much research and experimental results that have shown
that multipath coherence times can be difficult to predict and will depend on the day, the
location, and the depth of communication link with results of worst-case being in the order of
seconds [44, 125, 134,137].
3.3.5 The Doppler Affect
In addition the Doppler effect is another source of time variability in an underwater
communication channel, which occurs when there is relative motion between the transmitter
and receiver.
The motion of AUV's relative to each other will cause two possible forms of Doppler distortion in
the received signal: Doppler Shifting caused by an apparent shift in frequency as the vehicles
move towards or away from each other and Doppler Spreading or its time domain dual
coherence time, which is the measure of the time-varying nature of the frequency
dispersiveness in the Doppler spectrum [109]. The Doppler shift (∆ fd) of a received signal is:
∆𝑓𝑑 = f ∆𝑆𝑐
(3.12)
where f is the original signal frequency and ∆S is the relative velocity between the moving
vehicles and c is the speed of sound in water.
As an example, if the vehicles were moving at a moderately slow speed of 1 m/s (2 knots)
relative to each other and the signal frequency was f = 40kHz then the Doppler shift would be
∆fd ≈ 27Hz. In addition, the Doppler spread or coherence time measurements, as mentioned
above, can be as long as 1 s. Thus Doppler shifting and spreading cause complications for the
receiver to track the time-varying changes in the channel which need to be designed into the
channel estimation algorithms and explicit delay synchronisation approach within
communication protocols. Modern underwater modem designs have made significant advances
in the sophistication of its signal processing mechanisms to manage Doppler and Multipath
issues. In addition the swarm formation algorithms in general attempt to maintain similar ranges
between vehicles, and similar relative speeds which will lessen the impact of Doppler effects
and thus will not be considered further.
3.4 Noise The ocean is a noisy place to operate in, which is especially problematic when trying to use
acoustic communication signalling; yet obtaining an accurate noise model is critical to the
calculations of SNR at the hydrophone. There are three major contributors to noise underwater:
47
Figure 3.7: Power Spectral density of the Ambient Noise; W (wind), S (shipping)
ambient or background noise of the ocean; self noise of the vehicle and intermittent noise
including biological noises such as snapping shrimp, ice cracking and rain.
Thermal noise is common to any electrical receiving system resulting from the thermal agitation
of its electrons and this will happen with any electrical receiving system. This noise, however,
will not be included in this work as the in-band contribution of thermal noise will be negligible.
3.4.1 Ambient Noise
Ambient Noise in the ocean has been well defined [134] and refers to the noise that occurs in
the ‘far-field’. It can be represented as Gaussian and having a continuous power spectrum
density (psd). For the frequency region of interest for AUV swarm communication systems (10
kHz to 100 kHz), the ambient noise psd decreases with increasing frequency, refer to Figure
3.7. The characteristics of the spectrum of ambient noise is made up of four components
(outlined below), each having a dominating influence in different portions of the frequency
spectrum, where f is in kHz, s is the Shipping Activity Factor, w is Wind State in m/s.
The empirical values of noise spectrum levels are given in dB re 1 μPa per Hz:
• Turbulence noise influences only the very low frequency regions f < 10Hz that includes
tides, waves and earth seismicity activity:
10logNturb(f) = 17 − 30log(f) dB re 1 μPa per Hz (3.13)
• Distant shipping noise dominates the 10 - 100Hz region and has a shipping activity
factor of s whose value ranges from 0 to 1 for low to high activity respectively:
10logNship(f) = 40 + 20(s−0.5) + 26log(f ) − 60log(f+0.03) dB re 1 μPa per Hz (3.14)
48
• Wave and other surface motion caused by wind and rain is a major factor in the mid
frequency region of 100Hz - 100kHz where wind speed is given by w in m/s:
10logNwind(f) = 50 + 7.5w1/2 + 20log(f ) − 40log(f+0.4) dB re 1 μPa per Hz (3.15)
• Thermal noise becomes dominant over 100kHz. The molecular thermal motion of the
sea occurs due to the thermal agitation of the molecules of water producing pressure
fluctuations at the face of the transducer. It is not until a frequency over 100kHz that the
thermal noise component begins to dominate and the overall noise psd begins to
increase, but this point moves further away from the frequencies of interest for AUV
communication particularly as the wind speed increases.
10logNthermal(f) = −15 + 20log(f) dB re 1 μPa per Hz (3.16)
Ambient noise has been shown to be 9dB higher in shallow water than deep water [20]. In
general, however, ambient Noise power will decrease with increasing depth due to the distance
from the surface and therefore the decline in shipping and wind noise as they become more
distant.
In Underwater Swarm Sensor Networking (USSN) as well as other underwater networking
operations, where AUV's will be working in relatively close proximity to each other, there will be
added an additional level of ambient noise to their operations due to the noise of the other
vehicles in the swarm, irrespective of the operating depth. As will be discussed in the next
section on Self Noise, the expectation is that this additional noise component will have a limited
impact on the acoustic communication in a similar way to the ‘Shipping Noise' component of
ambient noise, which does not have a major impact on frequencies above 10kHz that are used
for USSN communications.
3.4.2 Self Noise
Self-noise is defined as the noise generated by the vehicle itself as the platform for receiving
signals. This noise can reach the hydrophone mounted on the AUV either through the
mechanical structure or through the water passing over the hydrophone. The degree to which
turbulent flows cause transducer self-noise depends on the location (mounting) of the
transducer and its directivity characteristics [127]. Self-noise can also be seen as an equivalent
isotropic noise spectrum as presented by Urick [134] from work done during the Second World
War on submarines. In general, as with ambient noise, there is decreasing levels of self-noise
with increases in frequency however self-noise is also significantly affected by speed with
decreasing noise spectra when the vessels are travelling at slower speeds or are stationary [42,
78, 134].
Kinsler [78] notes that at low frequencies (<1kHz) and slow speeds, machinery noise dominates
and at very slow speeds self-noise is usually less important than ambient noise. Whereas at
higher frequencies (10kHz) propeller and flow noise begins to dominate and as speed increases
the hydrodynamic noise around the hydrophone increases strongly and becomes more
49
significant than the machinery noise. This is due to the cavitation from the propeller due to the
entrainment of air bubbles under or on the blade tip of the propeller. At higher speeds, self-
noise can be much more significant than ambient noise and can become the limiting factor.
The self-noise of different size and types of vehicles are as varied as there are vehicle designs
and there are few recent published values. Each vehicle itself produces large variations in self-
noise with speed and operating conditions [42]. Self-noise can be controlled by selection of
motor type, configuration, mounting and motor drivers. The trend for most AUV's will be the use
of small brushless DC electric motors which have been used on the development of the
SeaVision vehicle [86]. Preliminary testing of self-noise on these vehicles shows an increase in
noise due to increases in speed, as has been predicted, but there was no way in the simple
experimental set up to distinguish between machinery and hydrodynamic effects. Higher
frequency components (up to 20kHz) were present as the speed increased due to the increased
work from the thrusters. When the SeaVision vehicle hovered in a stationary position, the
frequency of the noise psd centred around 2kHz, which is out of band noise.
Holmes et al. [60] at WHOI recently investigated the self-noise of REMUS, their torpedo shaped
AUV, used as a towed array. At the maximum RPM of the AUV, the 1/3rd octave noise level,
when converted to source level by the calibrated transmission factor, was 130 dB re 1μPa at 1m
directly aft of the vehicle for a centre frequency of 1000 Hz. This would represent the radiated
noise source level for a vehicle moving at 3 knots (1.5m/s). Vehicles typically radiate less noise
in free operating conditions than in tethered conditions, so the second trail on the REMUS was
measuring the radiated noise of the vehicle to examine the power spectral density of the noise
as recorded on the hydrophone array as it was towed behind the vehicle. The results showed
the RPM dependent radiated noise in the aft direction at a distance of 14.6 m behind the vehicle
looking at frequency range up to 2500Hz which is out of band noise.
We argue here that in a closely operating swarm of AUV’s that there will be an accumulation of
noise related to this self-noise of each vehicle in the swarm that will be in addition to the ‘far-
field’ shipping noise related to the machinery and propeller noise. Even though this becomes
less significant at higher frequencies, there will be some contribution. Furthermore, the
hydrodynamic noise around the hydrophone of even slowly moving vehicles will also contribute
to some additional noise that will be referred to as Swarm Noise.
3.4.3 Intermittent Sources of Noise
The sources of intermittent noise can become very significant in locations or times that they
occur close to operating AUV swarms. The two major areas where research has been
undertaken are in the marine bio-acoustic fields and also the effect of rain and water bubbles
created by raindrops.
Major contributors to underwater bio-acoustic noise include:
• Shellfish - Crustacea - most important here are the snapping scrimp who produce a
broad spectrum of noise between 500Hz and 20kHz
50
• Fish - toadfish 10 to 50Hz
• Marine mammals - Cetacea - porpoises 20 to 120Hz
Rain creates a different noise spectrum to wind and needs to be dealt with separately as it is not
a constant source of noise. Urick [134] showed examples of increases of almost 30dB in the 5
to 10kHz portion of spectrum in heavy rain, with steady rain increasing noise by 10dB or sea
state equivalent increase from 2 to 6. Eckart [42] presented average value of rain at the surface
from 100Hz to 10kHz of -17 to 9dB. These main contributors to intermittent sources
predominate in the lower frequency ranges up to 20kHz. Thus, interference in the operating
frequencies of communication data signals is considered low.
3.5 Short-range Acoustic Communication Physical Layer Parameters This section will establish the physical layer parameters used for a short range USSN. In
particular, to established and model the Signal-to-Noise Ratio (SNR) characteristics expected at
the hydrophone on an AUV using the propagation loss and noise characteristics found above. A
brief investigation into the various possible modulation schemes used in underwater modems
will be followed by the determination of the bit error rate (BER) which will provide a realistic
representation of the channel for the testing and analysis of the new MAC protocol designs. The
unique work presented here is the characteristics of short-range communication, which has had
very little attention in underwater communication research.
3.5.1 Signal-to-Noise Ratio
The narrowband Signal-to-Noise-Ratio (SNR) observed at the receiver, based on the
assumption of no multipath or Doppler losses is given by:
𝑆𝑁𝑅(𝑟, 𝑓,𝑑, 𝑡,𝑤, 𝑠,𝑃𝑡𝑥, 𝜂,𝐷𝐼) = 𝑆𝑃𝐿𝑝𝑟𝑜𝑗𝑒𝑐𝑡𝑜𝑟(𝑃𝑡𝑥𝜂𝑡𝑥𝐷𝐼𝑡𝑥)
𝑃𝐿𝑜𝑠𝑠(𝑟,𝑓,𝑑,𝑡)∗𝜂𝑟𝑥∗𝐷𝐼𝑟𝑥∗∑𝑁𝑜𝑖𝑠𝑒(𝑓,𝑤,𝑠)∗𝐵 (3.17)
where B is the receiver bandwidth, SPLprojector the projector signal intensity, PLoss total
attenuation and Noise being the sum of all the noise terms established in Section 3.4.
3.5.2 Frequency Dependent Component of SNR
Signal Frequency was shown to have a major influence on both propagation loss and noise,
hence taking the frequency dependent portion of the SNR from Equation 3.17, as presented by
Stojanovic [125] for long-range communication, will provide a view on optimal signal frequency.
The frequency dependent component is inversely proportional to SNR, thus:
𝑆𝑁𝑅(𝑓) = 1𝑃𝐿𝑜𝑠𝑠 (𝑟,𝑓,𝑑,𝑡)∗∑𝑁𝑜𝑖𝑠𝑒 (𝑓,𝑤,𝑠)
(3.18)
Figure 3.8 (a) reproduces Stojanovic’s results for longer ranges and shows various ranges up to
10km using Thorp’s absorption model (spherical spreading) which has been used by several
authors ([23, 91, 118, 125]. This work has extends these results to investigate consequences
and optimum signal frequencies for shorter ranges of 100 m and 500 m. In addition, to illustrate
the variation between the Thorp and Fisher & Simmons absorption loss models developed in
Section 3.3.3. The results for shorter ranges are presented in Figure 3.8(b).
51
These figures show that there is a signal frequency where the frequency dependent component
of SNR is optimised, assuming that the projector parameters, including transmitter power and
projector efficiency, behave uniformly over the frequency band. The black dot at the apex of
each of the curves in Figure 3.8(b) indicates this optimum signal frequency for that range using
Thorps Absorption Coefficient. The two absorption models present similar responses and
optimum frequencies for both ranges with much smaller variation in optimum frequency at 100
m due to the significantly lower contribution made by the absorption coefficient of propagation
loss. As with Stojanovic’s results, as the range reduces the optimum signal frequencies
increases which is due to the reducing noise levels at higher frequencies in the ranges
investigated.
The impact that the Ambient Noise component has on the optimal signal frequency is seen most
dramatically at the 100 m range when the wind state is changed from 0 to 2 m/s, the optimal
signal frequency changes from 38 kHz (blue curve) to 68 kHz (green curve). From a
communication perspective, if two vehicles were operating at 100 m and 38kHz and the wind
state changed from 0 m/s to 2 m/s, there is a reduction of 9dB in the frequency dependent
component of SNR. This is not an absolute reduction in SNR as the projector parameters and in
particular the transmitter power level have not been considered here. It does however indicate
the significant impact that wind and wave action can play with data communication underwater,
and in addition this reduced SNR value does not include any increased losses associated with
the increase scattering that wave action can generate. The impact of shipping, found in the
Ambient Noise term, is not included here as its effect is minor on signal frequencies above
10kHz.
(a) Long range [125] (b) Short range with channel parameter variations
Figure 3.8: Frequency Dependent Component of Narrowband SNR
52
(a) Comparison of Propagation Loss Parameters (b) Comparison with Noise Parameter (wind)
Figure 3.9: Optimum Signal Frequency based on Optimising SNR (determined from frequency-dependent component of narrowband SNR)
In Section 3.3.3, Figures 3.3 & 3.4(b) showed that temperature variations had the most dramatic
impact on the signal frequency (up to ~80kHz) in the signal frequency ranges of interest. Figure
3.8(b) demonstrates this impact with the optimum signal frequency at 500 m changing from
28kHz when water temperatures are 0°C to 33kHz at 27°C. This is further explored in Figure
3.9(a) that investigates the changes in optimum signal frequency with range and environmental
parameters associated with propagation and noise losses. The impact of changes in range can
be seen if the vehicles moved from 100 m to 500 m (at wind state 0 m/s), the optimum signal
frequency to maintain highest SNR decreases from 38 kHz to ≈ 28 kHz, Figure 3.8(b) & 3.9(b).
Figure 3.9(a) and (b) show the optimum signal frequency verses range up to 500 m for the
various propagation loss and noise parameters; that is temperature and depth, within the Thorp
and Fisher and Simmons Absorption Loss models and wind in the Ambient Noise model. The
optimum frequency, decreases with increasing range due to the dominating characteristic of the
absorption loss. It can be seen in Figure 3.9(a) that as the range increases there is an
increasing deviation between the two models and between the parameters within the Fisher and
Simmons model but minimal change at the working ranges of AUV swarms of less than 100 m.
When wind is included, as in Figure 3.9(b), there is a dramatic change in optimum signal
frequency at very short ranges and this difference reduces substantially over the range shown.
This is due to the increasing significance of the Absorption Loss term relative to the constant
Ambient Noise term (as it is not range dependent), which reduces the effect of the Noise term
and therefore the wind parameter. In both Figure 3.9(a) and (b), the Fisher and Simmons model
provides higher optimum frequencies due to the more accurate inclusion of the relaxation
frequencies of boric acid and magnesium sulphate.
Based on the results and considerations above, and that the application of a USSN may
operate at shorter ranges than 100 m, a 40kHz signal frequency is used as the carrier
53
frequency for further modelling. This selection is further support by the fact that readily available
commercial modems operate at this frequency indicating utility in real-world operations [138].
3.5.3 Channel bandwidth
Having established that at different ranges there is an optimum signal frequency that provides a
maximum SNR, and assuming constant transmitter power and projector efficiency; there is an
associated channel bandwidth with these conditions for different ranges. To determine this
bandwidth a heuristic of 3dB around the optimum frequency is used. In a similar approach to
Stojanovic [125], the bandwidth is calculated according to the frequency range using ±3dB
around the optimum signal frequency fo(r) which has been chosen as the centre frequency.
Therefore, using Equation 3.18, the fmin(r) is the frequency when the frequency dependent
component of SNR at optimum frequency SNR(fo) minus SNR(fmin) is ≥ 3dB, thus:
PLoss(r, d, t, fo(r))*N(fo(r)) − PLoss (r, d, t, fmin))*N(fmin) ≥ 3dB holds true (3.19)
and similarly for fmax(r):
PLoss(r, d, t, fmax))*N(fmax) − PLoss (r, d, t, fo(r))*N(fo(r)) ≥ 3dB holds true (3.20)
The system bandwidth B(r,d,t) is therefore determined by:
B (r,d,t)= fmax(r)−fmin(r) (3.21)
Thus, for a given range, there exists an optimal frequency from which a range dependent 3dB
bandwidth can be determined as illustrated in Figure 3.10. The changes discussed in Section
3.5.2 related to changes in the optimum signal frequency with changes in range and channel
conditions such as temperature, depth and wind. These variations are reflected in a similar
manner to the changes seen here in channel bandwidth and in turn will reflect in the potential
data transmission rates. Figure 3.10 demonstrates that both the optimal signal frequency and
the 3dB channel bandwidth decrease as range increases. The impact of changing wind
conditions on channel bandwidth is significant, however as discussed, wind and wave action will
also include time variant complexities and losses not included here. Temperature increases
show an increase in channel bandwidth, at ranges of interest, due to the reduction in absorption
loss as temperature increases, which means there are benefits in working in the surface layers.
The discussion here highlights that the underwater acoustic channel is severely band-limited
and bandwidth efficient modulation will be essential to maximise data throughput and essentially
that there is improved capacity potential at shorter ranges which also benefits multi-hop
arrangements [23].
54
Figure 3.10: Range dependent 3dB Channel Bandwidth shown as dashed lines. The Y-
axis is the Optimum SNR based on the frequency dependent component of the narrowband SNR
3.5.4 Theoretical Channel Capacity
Prior to evaluating the more realistic performance of the underwater data communication
channel, the maximum achievable error-free bit rate C for various ranges of interest will be
determined using the Shannon expression that considers an AWGN (Additive White Gaussian
Noise) channel for SNR calculations, Equation 3.22. In these channel capacity calculations, all
the transmitted power Ptx is assumed to be transferred to the hydrophone except for the losses
associated with the deterministic Propagation Losses developed earlier.
The Shannon expression using the Signal-to-Noise ratio, SNR(𝑟, 𝑓,𝑑, 𝑡,𝑤, 𝑠,𝑃𝑡𝑥, 𝜂,𝐷𝐼), defined in
Equation 3.17, is:
𝐶𝑐 = 𝐵 𝑙𝑜𝑔2�1 + 𝑆𝑁𝑅(𝑟, 𝑓,𝑑, 𝑡,𝑤, 𝑠,𝑃𝑡𝑥, 𝜂,𝐷𝐼)� (3.22)
where Cc is the channel capacity in bps and B is the channel bandwidth in Hz
Thus using the optimum signal frequency and bandwidths at 100 m and 500 m found in Section
3.5.2, the maximum achievable error free channel capacities against range are shown in Figure
3.11. The signal frequency and channel bandwidth values for 100 m were fo = 37kHz and
B=47kHz and for 500 m were fo = 27kHz and B=33kHz from Figure 3.10. These are significantly
higher than values currently available in underwater operations (Walree, 2007), however they
provide an insight into the theoretical limits. Two different transmitter power levels are used, 150
dB re1μPa which is approximately 10mW and 140 dB re1μPa is 1mW from Equation 3.1. Figure
3.11 illustrates the maximum theoretical channel capacities using these values and illustrates
that at the calculated optimum frequency and bandwidth for a range gives the highest channel
capacities which correlates with expectation.
55
Figure 3.11: Theoretical Limit of Channel Capacity (kbps) versus Range
The change in transmitter power, however, by a factor of 10, does not produce a linear change
in channel capacity across the range. These variations are important to consider, as minimising
energy consumption will be critical for AUV operations, however is out of the scope of this work.
In general, current modem specifications indicate possible data rate capacities of less than
10kbps [137] for modem operations under 500 m, well short of these theoretical limits.
3.5.5 Receiver (Hydrophone) Signal Intensity
To determine the hydrophone signal level, the total attenuation losses PLloss Equation 3.11 is
reduced from the starting projector signal intensity, SPLprojector Equation 3.5. Using a transmitter
power (Ptx) of 1 W gives a projector signal intensity of 168 dB re1μPa and a Ptx of 0.5 W is
164.8 dB re1μPa with a projector efficiency of 50%.
The sound signal intensity at the hydrophone (receiver) in dB re 1μPa is illustrated in Figure
3.12 based on a Ptx of 0.5 W and 1 W and a transducer efficiency of 50% and 30% for a range
of 20 to 50 m. Reductions in both Ptx and transducer efficiency see a deduction in received
signal intensity as is expected due to the reduced SPLprojector. The figure also illustrates the
reduction over range due to the attenuation from path loss.
56
3.5.6 Signal-to-Noise+Interference- Ratio (SNIR)
To establish the Signal to Noise + Interference Ratio, that will be used in the non-ideal
simulations in Chapter 6, an interference level (FL) at the receiver is included with the SNR of
Equation 3.17 to obtain:
𝑆𝑁𝐼𝑅(𝑟, 𝑓,𝑑, 𝑡,𝑤, 𝑠,𝐹𝐿,𝑃𝑡𝑥, 𝜂,𝐷𝐼) = 𝑆𝑃𝐿𝑝𝑟𝑜𝑗𝑒𝑐𝑡𝑜𝑟(𝑃𝑡𝑥𝜂𝑡𝑥𝐷𝐼𝑡𝑥)
𝑃𝐿𝑜𝑠𝑠(𝑟,𝑓,𝑑,𝑡)∗𝜂𝑟𝑥∗𝐷𝐼𝑟𝑥∗((∑𝑁𝑜𝑖𝑠𝑒(𝑓,𝑤,𝑠)∗𝐵)+𝐹𝐿) (3.23)
where FL includes the reverberation level, RL that includes the time delayed self interference
signals from multipath and any packet interference that may occur at the receiver. At this stage,
it is assumed that interference due to other vehicle packets arriving during reception is not
included with only the self-interference and reverberation levels included as discussed in
Section 3.3.4. Matlab code is available in Appendix B.
Figure 3.13 illustrates the declining SNIR (at f = 40 kHz) with range over the ranges of interest
in swarm operations and the impact that changes in the transmitter power (Ptx), transducer
efficiency (Eff = ηtx = ηrx) and Reverberation Level (RL) will have on the SNIR. Changes in
ambient noise are insignificant due to the dominance of the reverberation level interference as
is supported by Urick [134] and Waite [137]. The impact that wind and therefore sea states has
on the SNIR, which is recognised as having a strong influence on ambient noise at 40 kHz
(Section 3.4), has an indirect impact through the reverberation levels that will be varied
dynamically in the simulations conducted in Chapter 6 to reflect these conditions.
Figure 3.12: Receiver Signal Intensity vs Range for Variation in Transmitter Power and Transducer Efficiency
20 25 30 35 40 45 50120
125
130
135
140
Range (m)
Rec
ieve
d S
igna
l Int
ensi
ty d
B re
1uP
a
Ptx=1W, Eff=50%Ptx=1W, Eff=30%Ptx=0.5W, Eff=50%Ptx=0.5W, Eff=30%
57
Figure 3.13: SNIR vs Range for variation in Transmitter Power, Transducer efficiency, and Reverberation Level
Increases in electrical power at the transmitter can be seen to support the reduction in bit error
rates that are a result of higher SNIR. The transmitter power levels are significantly influenced
by the efficiency of the transducer as can be seen by the reduction in SNIR with a change of
efficiency from 50% to 30%. The almost 5 dB change due to the efficiency change occurs above
the expected 2dB (10log(0.5/0.2) due to the non linear characteristics resulting in the channel.
The additional power in the transmitted signal due to better efficiencies will create increased
reverberation levels. Minimising transmitter power will be beneficial in the reduction of overall
energy consumption from the communication requirements of a swarm vehicle.
3.5.7 Modulation and Bit Error Rate (BER)
Achieving close to the maximum channel capacities as calculated in Section 3.5.4 is still a
significant challenge in underwater acoustic communication systems due to the harsh channel
conditions for communication as discussed in this chapter and seen in the SNIR. Depending on
modulation and error correction schemes, the probability of bit error, BER, can provide a
measure of the data transmission link performance.
In underwater systems, the use of FSK (Frequency Shift Keying) and PSK (Phase Shift Keying)
have occupied researchers’ approaches to symbol modulation for several decades. One
approach is using the simpler low rate incoherent modulation frequency hopping FSK signalling
with strong error correction coding that provides some resilience to the rapidly varying multipath.
Alternatively, the use of a higher rate coherent method of BPSK signalling that incorporates a
Doppler tolerant multi-channel adaptive equalizer has gained in appeal over that time (Johnson
et al., 1999).
20 25 30 35 40 45 50-5
0
5
10
15
20
Range (m)
SN
IR (d
B)
Ptx=2W, Eff=50%, RL=-45dBPtx=1W, Eff=50%, RL=-45dBPtx=0.5W, Eff=50%, RL=-45dBPtx=1W, Eff=30%, RL=-45dB
58
The BER formulae are well known for FSK and BPSK modulation techniques [109], which
require the ratio of Energy per Bit to Noise psd, 𝐸𝑏𝑁𝑜
, that can be found from the SNIR (Equation
3.23) by:
𝐸𝑏𝑁𝑜
= 𝑆𝑁𝐼𝑅(𝑟, 𝑓,𝑑, 𝑡,𝑤, 𝑠,𝐹𝐿,𝑃𝑡𝑥, 𝜂,𝐷𝐼) ∗ 𝐵𝑐𝑅𝑏
(3.24)
where R is the data rate in bps and Bc is the channel bandwidth. Equations 3.25 and 3.26 are
the uncoded BER for BPSK and FSK respectively:
BPSK: 𝐵𝐸𝑅 = 12𝑒𝑟𝑓𝑐 �𝐸𝑏
𝑁𝑜�12� (3.25)
FSK: 𝐵𝐸𝑅 = 12𝑒𝑟𝑓𝑐 �1
2∗ 𝐸𝑏𝑁𝑜�12� (3.26)
A channel bandwidth of Bc = 5kHz and data rate R = 9.6 kbps will be used in the following
analysis based on the discussions above and a reflection of the current maximum commercial
achievable levels examined in Section 3.5.7.1. Modulation curves showing BER for 𝐸𝑏𝑁𝑜 can be
found in many textbooks [83,109] for both BPSK and FSK modulations. Figure 3.14 illustrates
the BER against ranges between vehicles using the SNIR of Equation 3.23.
From the BPSK modulation table, given at Appendix C and taking a BER of 10−4 or 1 bit error in
every 10, 000 bits, the minimum 𝐸𝑏𝑁𝑜
required is 8.4 dB. This is an effective SNIR, which using
Equation 3.24 gives the SNIR of (8.4 + 0.2 dB) 8.6 dB using the channel bandwidth and data
rate given above.
Figure 3.14: BER vs Range for Short Range Acoustic Data Transmission Underwater
20 25 30 35 40 45 5010-10
10-8
10-6
10-4
10-2
100
Range (m)
BE
R
BPSK: Ptx=1W, Eff=30%, RL=-45dBFSK: Ptx=1W, Eff=50%, RL=-45dB BPSK: Ptx=1W, Eff=50%, RL=-45dBBPSK: Ptx=2W, Eff=50%, RL=-45dBBPSK: Ptx=1W, Eff=50%, RL=-50dBBPSK: Ptx=1W, Eff=50%, RL=-55dB
59
Unless otherwise stated the curves of Figure 3.14 are based on a RL of -45 dB [137], which is
being used as an average ‘interference’ level. If this value reduces by just 5 dB the range in
which no bit errors will occur, using BPSK, increases from only 24 m between vehicles to being
able to handle up to 40 m. Transmitter power also plays a critical role. Take for example BPSK
with Ptx of 0.5W (the same curve as the FSK at 1W) has a BER higher than 10−4 for all ranges
shown and therefore cannot be used. At 1 W the swarm vehicles would only be able to operate
up to ~24 m before bit errors are likely. Increasing the transmitter power from 1 W to 2 W
increases the range from 24 m to 33 m. To reach 50 m without bit errors, the transmitter power
or transduce efficiency would need to increase under the proposed noise and interference
conditions. This will be explored further with the simulations in Chapter 6.
3.5.7.1 Currently Available Acoustic Modem Capacities
There are very few commercially available acoustic modems, designed and built for short-range
sensor networks. Wills et al [144] have begun a project specifically for this purpose, however no
sea trials have been done and therefore there are no published details on its capabilities.
Parrish [100] used the range vs rate product to compare telemetry systems in 2000. For short
range applications they found two commercially available modems; one for shallow water rated
at approximately 1000 baud at 600m, to deep water providing 40 to 100kbps at 100m. Whereas
Vasilescu et al (MIT/CSIRO) [135] required analysis on 200m and assumed 480bits/s (at
4.5mJ/bit) based on Aquacomm modem.
This illustrates the incredibly severe data communication environment found underwater and
that commercial modems are generally not yet designed to be able to adapt to specific channel
conditions. In addition there are technological parameters worth noting that mean that it is
easier to transmit at higher data than it is to receive on an AUV. This is because of the
additional need to deal with propulsion noise on reception as well as the difficulties in mounting
receiver arrays on small AUV [101].
3.5.8 Long-Range vs Short Range
Short-range underwater communication systems have three key advantages over longer-range
operations: the delay between transmitter and receiver will be lower; the signal attenuation is
smaller and less dependent on channel conditions and the optimum signal frequency and
bandwidths are higher [3, 6]. The delay at 30 m for example is approximately 0.02 s which is
considerable lower than the 2 s delay at 3 km but still significantly higher than is experienced in
RF communication.
The smaller signal attenuation means lower transmitter power requirements, which will result in
reduced energy consumption, which is a factor for AUVs that rely on battery power. Battery
recharge or replacement during a mission is difficult and costly, although the proportion required
for communication is small in comparison to the thruster control (Appendix D) for movement.
Attenuation will be less, but more significantly, there will be less weighting on the absorption
component of attenuation, which means less dependency on temperature, salinity and depth
60
(pressure). This also signifies less emphasis on frequency as the frequency dependency part of
attenuation is in the absorption component and thus will allow the use of higher signal
frequencies and higher bandwidths at short ranges.
Reverberation, which was discussed in Section 3.3.4.2, is a significant challenge for underwater
acoustic communication due to the resultant multipath signals that can lead to Inter Symbol
Interference (ISI). In terms of its impact due to range, shorter-range communication may be less
problematic than longer-ranges due to less boundary reflections, especially if the range is much
less than the depth [134] and the difference in the various path lengths is lower in proportionate
terms for longer transmission range.
A new type of reverberation has been hypothesised, Section 3.3.4.1, and presents a
disadvantage of multiple vehicles operating a very short-range that will cause interference due
to reflections of sound signals off the vehicles themselves. That is, with many swarm vehicles
operating in close proximity with each other and utilising frequent packet transmissions that
there will be a substantial amount of reverberation from these data transmissions in the vicinity
of each vehicle’s hydrophone. It is proposed that it will insert a significant interference
component into the noise level at the receiver and will be developed as a SNIR in the non-ideal
simulations presented in Chapter 6. This reverberation therefore can become a challenge for
high-density short-range swarm operations.
Reverberation and noise must be jointly considered as background for the consideration of
detection of a signal with the determination of which contributor is most responsible for creating
the interfering background [44]. Ambient ocean noise is not affected by range however the
impact of neighbourhood vehicle noise due to densely operating AUV’s does become more
significant as range between vehicles becomes shorter.
Table 3.2: Comparison of Terrestrial and Long and Short range Acoustic Bandwidths
Terrestrial
Underwater
Short Range Long Range
Wave Type electromagnetic acoustic
Velocity 3x108 ms-1 1500 ms-1
Propagation Delay micro seconds milliseconds seconds
Frequencies 500kHz-300GHz 30-100kHz 10 – 30kHz
Energy Consumption low important critical
Transducer electromagnetic piezoelectric / half duplex
Receiver less complex complex
61
3.6 Conclusion The performance of acoustic data communication underwater is characterised by high and
variable propagation delays, propagation losses that are both range and frequency dependent
and background noise that is frequency dependent. Some of these extreme features are
somewhat improved for shorter range communication as discussed in this chapter, yet are still
significantly poorer than are experienced in RF communication. These differences with RF
communication call for different approaches to the design of underwater communication
protocols with even further refinement required when dealing with shorter ranges.
In the application of swarming networks, which has many agents in a dense topology, places
more demand on equal access to the same medium and therefore new constraints on
particularly the MAC layer design, and this will be investigated further in the next chapter. In
addition, the channel geometry that can create severe multipath fading
This chapter has presented a model of the short-range underwater acoustic channel and
analysed the important properties and transmission channel state information for the design and
development of a communication protocol for swarming AUV networks. To effectively design
these protocols it is necessary to understand these properties to optimize QoS requirements in
the different layers of the protocol stack.
62
This page intentionally blank
63
Chapter 4 Medium Access Challenges for Underwater Swarm Sensor
Networks
4.1 Introduction Despite the intensive research over many decades that has occurred into Medium Access
Control (MAC) protocols for wireless ad-hoc and sensor networks for terrestrial systems, the
study of MAC protocols for underwater networks has begun much more recently. Even so it is
rapidly evolving. The focus on communication protocols however for an underwater
autonomous swarm of AUVs is still in its infancy. The focus of this Chapter is to analysis the
current state of the art and related works associated with underwater acoustic MAC protocols
required for acoustic ad-hoc sensor networks and to highlight the developments and challenges
in mobile underwater multi-vehicle swarm networks.
Accessing the underwater acoustic channel presents different and even more challenging
physical conditions than in wireless terrestrial systems, as discussed in Chapter 3, and
therefore requires new approaches to communication protocols and MAC layer designs [4, 32,
26, 88, 99]. It has been established in Chapter 3 that there are some advantages when
considering designing shorter-range communication, which are required for swarm networks,
compared to the more typically used longer range acoustic systems. In particular, shorter-range
networks can utilise higher frequencies (albeit still in the mid kHz range), lower propagation
delays (but still in the milliseconds to second range) and have lower path losses to deal with
which are less dependent on absorption losses (sea water conditions). Multipath spreads, even
in shallow water will be less severe as range difference from the direct path to the reflected
paths will be much smaller and angles of incidence of reflected signals are such that they will
not travel towards the receiver.
The very slow propagation speed of an acoustic signal (compared with RF) and the 3D
distribution of vehicles create a space-time diversity that is unique to underwater acoustic
communication and is very different to RF networks. This phenomenon, in particular, impacts on
the MAC layer timing of transmissions and its knowledge of the start and end of packet frames.
As discussed in Chapter 2, the communication protocol is to be designed for a closely operating
group of AUVs functioning in a single-hop swarm where the 3D structure of the swarm attempts
to maintain vehicles operating at similar distances in the order of 30 to 50 m. The swarm is
made up of a group of simple low cost vehicles operating in a decentralised and distributed
network that require continuous communication of localisation and mission control information.
The communication between vehicles needs to support the swarm so that it can be self-
configuring and scalable to deal with device failures and losses while being self-correcting in
dealing with the harsh channel conditions and lost data packets. The main feature required of
64
the protocol is to exchange information from every vehicle to every other vehicle in the swarm
with time-sensitive data constraints to maintain swarm synchronisation (preserve its formation)
and to avoid vehicle collisions. Thus for a USSN, the QoS requirements are for delivery of time-
sensitive data exchange and reliability for dependable and consistent network connectivity.
This Chapter begins with a review of the current knowledge and research of MAC designs for
an Underwater Wireless Sensor Network (UWSN) with a focus on mobile Underwater Swarm
Sensor Network (USSN) and will provide an examination of the protocol structures and
challenges that are influential in the design of a medium access protocol for an underwater
swarm network. This leads to a discussion of the unique spatial-temporal diversity that is
experienced underwater when using acoustic signalling and its impact on the MAC design.
4.2 MAC Protocol Overview The main purpose of the Medium Access Control (MAC) protocol layer is to provide addressing
and coordinated channel access control mechanisms (such as scheduling, buffer management
and error control) that make it possible for multiple users to share a common communication
channel. Accessing the communication medium needs to be controlled to limit or prevent packet
collisions as this is extremely problematic underwater due to the low bandwidths available and
the latency of propagation of acoustic signals not only for transmission but also if retransmission
is required. Retransmissions are doubly wasteful due to the need to retransmit packets multiple
times and due to latency associated with informing the need to retransmit. In the harsh
environment for communication underwater the MAC protocol needs to exploit the key
properties of the channel (low bandwidths and high latency) to maximize the channel utilization
and throughput. An additional major focus of MAC algorithms is to limit any wasteful energy
consumption as communication consumes the majority of a wireless sensor node’s energy. As
discussed in Chapter 3 this is not a major concern in USSN, due to the energy consumption of
the communications being substantially less than the propulsion system and missions are
generally short, and thus discussion here will not focus on these underwater MAC
developments.
MAC protocols can be designed to allocate transmission resources either in a fixed or in a
dynamic manner. Fixed channel allocation techniques such as Frequency Division Multiplexing
(FDM) or Time Division Multiplexing (TDM) are commonly used in many communication
systems where ample channel capacity is available to transmit information [74]. Channel
allocation techniques have been developed based on two broad categories; Random Access
(generally contention based) or Scheduled (mainly contention free) [73]. Scheduled protocols
such as polling can be modified to produce an adaptive polling protocol where resources can be
allocated dynamically. Also in a polling protocol the transmitted resources can be adjusted
dynamically in every cycle.
MAC algorithm designs can be further broken down into single-channel or multi-channel and
they support centralized or decentralised network topologies. For low data rate and variable
65
channel conditions, dynamic channel allocation techniques are generally used to maximise the
transmission resources.
4.2.1 Random Access
A pioneering random access protocol for wireless packet data networks is ALOHAnet or simply
ALOHA [1] that essentially uses uncoordinated access to a medium. The first version, ‘Pure
ALOHA’, was very simple with the uncomplicated rules of; if you had data to send, then send it
immediately; and if the packet collides with another transmission (which becomes known to the
sender if it does not receive an acknowledgement (ACK)), then try resending at a random time
later. The quality of the mechanism to determine the random ‘later’ time significantly influences
the efficiency of the protocol [73]. Other major variations of the ALOHA protocol include Slotted
ALOHA [110] and a Reservation ALOHA scheme [30]. Analysis of the probability of collision has
lead to determining the maximum normalised throughput achievable in the Pure ALOHA and
Slotted ALOHA protocols of 0.184 and 0.368 respectively. Interestingly, Ahn [3] determined that
in long and variable propagation delay environments such as underwater, the throughput of
slotted ALOHA degrades to pure ALOHA, which means that even slotted ALOHA delivers poor
throughput underwater.
It is not only this low achievable throughput but also the large energy wastage due to packet
collisions that are major issues for adapting this protocol for use underwater. It does have
several advantages such as no control packets, no time synchronisation requirements and no
sensitivity to channel variations [34, 83] however unless throughput can be improved, it does
not present good outcome for USSNs.
Chirdchoo [24] proposed two enhanced ALOHA schemes to take advantage of the long
propagation delays in underwater environments: ALOHA_CA (Collision Avoidance) and
ALOHA_AN (Advance Notification). The novel approach of ALOHA_CA is for each node to pay
attention to every packet that it overhears and record sender and receiver information.
ALOHA_AN has in addition to ALOHA_CA a control packet NTF (advanced notification packet)
that gives the sending nodes some indication of better times to transmit. ALOHA_CA is simpler
and more scalable than ALOHA_AN however it was demonstrated that ALOHA_AN could
achieve higher throughputs and higher than Slotted ALOHA. Ahn [3] proposed a Propagation
Delay Tolerant (PDT)_ALOHA that added guard times to slotted ALOHA increasing throughput
by 17% to 100% depending on traffic conditions.
The most popular choice of Random Access protocol in RF sensor networks is the CSMA
(Carrier Sensing Multiple Access) based protocols such as CSMA/CA [147]. The principle of this
protocol is that a transmitter that wants to initiate a transmission checks the transmission
channel by checking the presence of a carrier signal. If no carrier signal is present, which
indicates the channel is free; the transmitter will initiate a transmission. These protocols are
therefore good for scalability and adaptability and there is no need for time synchronisation that
makes them well suited for distributed, ad-hoc networks such as IEEE802.15.4 [63]. However,
despite these advantages, which would appear to be beneficial for underwater swarm
66
operations, it has been shown that with high propagation delay environments random access
approaches reduces the efficiencies of the CSMA process and reduces throughput [57, 65] due
to packet collisions and the need for retransmissions. Due to the spatial-temporal diversity
experienced in high propagation delay environments these protocols also are affected by the
hidden and exposed terminal problems.
To examine these phenomena in an underwater scenario, take the network example shown in
Figure 4.1. Assume firstly a point-to-point link between Nodes A and C that is 150 meters and
Node A starts transmission of a packet. The propagation delay for a 150 m link underwater is
0.1 s (using the speed of sound as 1500 m/s). If Node C senses the channel during the
propagation delay time, then it will also find the channel is free and could start transmission. In
this case both packets could collide and the transmission channel capacity will be wasted for a
period of at least the packet transmission time and the expected propagation delay. In
comparison, an RF channel with a 150 m link will incur a propagation delay of only 0.5 μs, that
is about 200,000 times shorter than for an acoustic delay, and therefore can be considered
negligible.
The throughput on such an acoustic link will be impacted by the ratio of packet length to
propagation delay. Take for example Node A transmitting a 100 byte packet to Node C, then the
packet will take about 0.08 s to transmit on a 10 kbps RF link. The same packet will suffer a
0.18 s propagation delay (assuming 10 kbps acoustic link) offering a net throughput on the
acoustic link of 4.4 kbps. This calculation is based on the assumption that the transmission
channel is ideal i.e. BER=0. If the BER of the channel is non-zero then the throughput will be
further reduced.
Figure 4.1: Hidden and Expose Node Problem
If D wants to tx to B it is ‘exposed’ due
to A’s transmission
A could suffer from near-far
effect, if D much closer B is ‘hidden’ from A’s
transmission
A
C
B
D A transmits
67
Figure 4.2: Minimum CSMA cycle with handshaking
Thus, a Random Access MAC approach on a point-to-point link that allows a user to access the
channel at a random time independently from the other users has a high potential for packet
collisions even when carrier sensing has been included in underwater operations. When there
are many nodes and several links in a network, packet collisions may occur due to the hidden or
exposed node problem and the ‘near-far’ effect. If Node A is transmitting to C then B is ‘hidden’
or cannot ‘sense’ Node A’s transmission and therefore will assume the channel is free. On the
other hand, Node D might want to transmit to B however it ‘senses’ or is ‘exposed’ to Node A’s
transmission to C. In addition, if Node A is much further away than say Node D then Node D’s
signal strength could be much higher and swamp A’s transmission which is referred to as the
‘near-far’ effect.
These potential packet collision scenarios have led to many random access protocols
developed that include collision avoidance such as ALOHA_AN (ALOHA_Advanced
Notification) [24] and Slotted FAMA [89]; handshaking approaches such as APCAP [53] and
2MAC [65]; and retransmission strategies, that rely on various back-off mechanisms such as
DACAP [102] and R-MAC [141]. A discussion of these and other protocols will be given
following an explanation of the challenges that handshaking mechanisms introduce in
underwater scenarios.
In underwater networks, any additional data that needs to be sent for control purposes (such as
handshaking) is a major liability not only in terms of the propagation delays, but also as the
near-far, hidden and exposed node problems are likely to be encountered for control packets as
well as the data packets themselves. To demonstrate the effect of handshaking in a propagation
delay environment and to consider the channel utilisation and cycle time of a CSMA protocol
with handshaking, Figure 4.2 illustrate this using values typical of an underwater network.
Assuming RTS & CTS packets of 20 bytes each and the data packet of 150 bytes, and a data
rate of 10 kbps, then the transmission times are 0.016 s & 0.12 s respectively. At a range
sending its data packet, to ensure avoiding collisions from possible hidden nodes, gives a cycle
time of : 2*0.016 + 2*0.1 + 0.2 + 0.12 + 0.1 = 0.652 s. Thus channel utilisation including
Node A (Sender)
Node B (Receiver)
Data
CT
RTS
Typical cycle
Wait
68
between transmitter and receiver of 150 m the propagation time is ~0.1s. Adding a wait time of
maximum propagation time (say 0.2 s at 300m), between the sender receiving the CTS and
RTS/CTS is ~18.4% (0.12/0.652) which is considerable less than would be seen in RF systems
and extremely poor exploitation of the channel. Thus, for random access protocols in
underwater networks, it is essential to minimise control data exchanges to improved channel
utilisation and minimise delay while recognising that some mechanism is required to avoid
packet collisions. These feedback based systems simple impact the timing of data exchange.
There have been many approaches suggested to improve the handshaking mechanism in
underwater protocols. These include trimming aspects of the control packets [33, 89, 129],
allowing packet trains [53, 89, 119, 143] and using reservation and back-off mechanisms [143,
22].
The size of the RTS/CTS packets were dramatically reduced by using just a tone in T-Lohi [129]
and in S-FAMA [89] by introducing time slots that imposed restrictions on when packets could
be sent which allowed the use of much shorter control packets. This helped lower the probability
of collisions, however throughput and packet delay remain limited as slot lengths were only
marginally reduced as they were still required to incorporate maximum propagation delay guard
times.
The 2MAC protocol [33], designed to support swarming of AUVs, has been based on MACAW
(Multiple Access with Collision Avoidance for Wireless) using the 4-way handshaking access
method RTS-CTS-DATA-ACK of MACAW plus adding a new control packet called BTS
(Blocked to Send), which is used to inform its neighbours that the channel will not be available.
2MAC showed considerably better throughput performance over a CSMA/CA protocol, which
was explained by its use of OFDMA technology at the physical layer. OFDMA permitting
multiplexing over a single frequency band, which allows the use of multiple channels with a
single transceiver, and was the major contributor to the reduction in collisions. The 2MAC
protocol did, however, improve the performance over the CSMA/CA protocol with the overall
throughput performance still very poor. In addition, it was recognised that the use of
handshaking meant the ETE [End-to-End] delay would be potentially very high, although not
specifically investigated.
Chen [22] proposed a bio-inspired MACAW approach where a leader requests a position report
from its followers who reply with data allowing the leader to create a new mapping of the
locations of followers. The leader uses this collected data to determine the next position of each
follower, which is then sent back to each follower who replies with an ACK. The topology was
based on a set of gliders that were operating on average at 1.6 kilometres between each other.
This means that the propagation delay is approximately 1.07 s ( 1600 𝑘𝑚1500 𝑚/𝑠
) and transmission times
ranging from 0.38 s to 3.2 s depending on modulation scheme and packet type required. Thus,
the time for a packet to be transmitted and successfully received would be between 1.45 s to
4.27 s assuming no errors. This would increase, of course, if retransmissions were required.
69
However, as vehicles are travelling at 0.25 m/s it would take almost an hour before vehicles
collide thus not requiring time critical communication.
Several protocols have suggested the use of packet trains that allow the scheduling of many
data transmissions with a single handshake to improve performance [53, 89, 119, 143] with the
idea that once the node had captured the channel it should empty its queue. This is sensible for
applications with bursty traffic and where data is delay insensitive however it can create unfair
allocation of the transmission channel among the users.
To reduce packet collisions, various reservation algorithms have been tested. In R-MAC [143]
reservations are made via a latency detection phase where a Neighbour Discovery Packet is
utilised to create a latency matrix in each node that provides a mechanism for a node to
determine its own schedule. Guo’s [53] APCAP (Adaptive Propagation Delay Tolerant Collision
Avoidance Protocol) modifies the RTS/CTS process by adding delay related to expected
propagation delay and then using the gaps between the RTS/CTS and data packets to schedule
other RTS/CTS and data packets. Both protocols showed improved throughput particularly as
load increased.
Gains have also been shown with various random back-off procedures to improve effectiveness
of the handshaking mechanism in high latency environments [39] and to more efficiently
allocate the timing of retransmissions after collisions [89] to minimise delay. In a different back-
off approach Peleato [102] included an additional warning signal during handshaking based on
knowledge of range between transmitter and receiver to reduce waiting time before sending the
data packet.
Noh [94] suggested taking away the handshaking approach altogether in their DOTS (Delay-
aware Opportunistic Transmission Scheduling) protocol, and instead use a passive overhearing
to build a location map of all vehicles so that it can determine its own transmission schedule in a
similar way to R-MAC. This will reduce overheads from a handshaking mechanism; however, in
highly mobile environments maintaining location maps becomes problematic due to the
constantly changing location of vehicles. Similarly, Molins [89] uses the knowledge that the
sending node has to determine if a packet might collide at the receiver and schedules time slots
accordingly in order to lower probability of collision.
The variations that have been designed into random access protocols to better fit the
underwater wireless networks show positive results for the particular applications that they have
been designed for. These applications, however, are focused on sensor networks that are
predominately stationary and where nodes are more sparsely deployed with lighter intermittent
or bursty traffic which means that data in node queues can be delivered when traffic eases and
packet trains can be accommodated. This is not the traffic condition expected in a swarm of
underwater vehicles. In fact, these contention-based approaches do not take advantage of the
knowledge of the continuous data traffic conditions required by the swarm.
For mobile underwater swarm sensor network operations, range between vehicles is much
shorter and larger packets and packet trains are unnecessary as the essential feature is to send
70
the most up-to-date data, which is time-sensitive. In fact, it is preferable to aggregate data in
some way to minimise packet size which also has the benefit of being less affected by multipath
underwater [125].
Thus, due to the close operating ranges of AUVs in a swarm, the network communication traffic
will be continuous and relatively heavy which requires a highly interconnected network
configuration [117, 114]. This regularity of packet generation and need for message exchange
tends to suggest the use of scheduled protocols. However, attention to some of the common
characters associated with scheduled protocols such as the need for time synchronisation,
limitations with scalability [131, 25] and lack of flexibility which are particularly necessary for
swarm communication due to the irregularities which come from both the mobility of vehicles
and the variability of underwater channel conditions.
4.2.2 Scheduled Protocols
The three fundamentally different contention free or scheduled multiple access approaches are
FDMA, TDMA and CDMA using frequency, time and codes respectively to partition the available
bandwidth and schedule multiple users. Polling and token ring protocols are also schedule-
based protocols and bring more flexibility to the scheduling and thus will be investigate further.
A major advantage of schedule protocols is that they can avoid packet collisions, which can
avoid energy wastage and additional delays due to retransmission. The disadvantages relate to
the inflexibility around the prearranged allocation of transmission resources.
As the bandwidth is already limited in underwater channels, dividing it further to create
frequency sub-bands in FDMA makes it particularly vulnerable to frequency fading as the
bandwidth of sub-channels may be smaller than the coherence bandwidth of the transmission
channel [121, 124]. Daladier [33] investigated the option of an OFDMA (Orthogonal Frequency-
Division Multiple Access) physical layer combined with a contention based MAC layer protocol
for a swarm of AUVs. It was determined that only 3 sub-channels were possible because of the
small bandwidths available underwater. The results showed that having 3 channels did reduce
packet collisions however, it resulted in very poor throughputs of less than 10% for all but the
lightest of offered loads.
CDMA as a MAC layer protocol has shown some potential in underwater networks as it provides
some resistance to multipath propagation and frequency-selective fading but requires high
receiver complexity, and has limitations as the number of nodes increase because it is difficult
to assign unique pseudo-random codes to a large number of nodes. Code reuse strategies
have been developed [102, 113] in a similar way to frequency band reuse in cellular networks,
however this becomes more complex in mobile environments.
In TDMA, time is divided into time-slots, which are often incorporated into repeating frames or
cycles and is the most common scheduled-based protocol. It is the variations in time-slot length
and time-slot allocation or scheduling that determine the nuances of different TDMA protocols.
Time slot lengths are determined by the transmission time of packets plus a guard time that is
incorporated to take into account the variable and large propagation delays that are possible
underwater and the potential time synchronisation errors. For slot allocation mechanisms,
71
options range from as simply using the ID number of the node to sequence node slots through
to variations based on position in network, which becomes more complex as mobility is
introduced.
Polling is the third major channel access mechanism for radio communication after TDMA and
CSMA/CA respectively [133] and is used in a wide variety of communication protocols in
terrestrial networks such as IEEE 802.4 token bus and 802.5 LAN [11, 73]. The features of a
polling based protocol sits somewhere between TDMA, which has control over the channel, and
CSMA/CA, which has flexibility by using random access. Polling uses a token to indicate when
a node has finished transmission or when it can begin transmissions.
Thus polling protocols incorporate a form of reservation to the TDMA cycle where instead of a
cycle being a fixed schedule based on a regular time for transmissions, the polling protocol
relies on the arrival of a token to trigger a transmission from the next node in the schedule. The
polling protocol has a flexible structure that can be adopted for different applications, however
fundamentally supports a centralised topology structure [2] but can also work with a
decentralised hierarchical [15, 43] architecture.
There are two fundamental operational modes of polling networks: roll-call polling or hub-
polling. Roll-call polling requires a central controller to control the token by initiating all
communication. Hub-polling is like a token-ring network where instead of the token having to
return to the central controller each time, it is passed on by each node. For long propagation
delay environments, such as underwater, roll-call polling is less efficient in that it becomes a
form of handshaking between nodes in the network as discussed in Section 4.2.1 and Figure
4.2. Hub-polling or the token-ring approach, that do not need the extra control signalling, appear
to offer a good compromise between the inflexibility of TDMA and the flexibility of CSMA/CA.
4.3 Time Scheduled Medium Access and Token Polling Approaches Using a time-based protocol has many advantages for swarm communications. Due to the
density of vehicles, the broadcast nature and continuous traffic required in swarm
communication, the probability of packet collisions occurring at the receiver is very high if a
random access based protocol is used. Thus, a scheduling protocol can be designed to avoid
packet collisions and therefore avoid the need for re-transmissions as well as avoiding the
issues of hidden and exposed nodes.
In addition, a TDMA approach has the advantage of simplicity, from a hardware and
computational perspective and energy efficiency viewpoint [144, 133], which are practical
features important to consider in any underwater system.
4.3.1 TDMA Based Protocols for Swarming AUVs
The concentration of effort around underwater multi-vehicle mobile networks has been on more
sparsely deployed glider and AUV networks and their co-ordination algorithms [22, 31, 98,103,
117, 144, 146], with less emphasis on the explicit communication between vehicles. In these
sparsely deployed networks, information exchange updates are generally only required on a
72
periodic basis and therefore more sporadic communication is acceptable, which has included
vehicles surfacing periodically [98, 102]. Petillo [103] suggested that until better swarm
communications is possible, then ‘Periodic Surface Communication' is the best option that
allows AUVs to surface with enough frequency to obtain information that could re-direct them to
more optimal sampling positions. Paley [98] also used periodic surfacing to access the RF
network and to access GPS co-ordinates. The option of surfacing however takes away the
ability of maintaining a practical real-time swarm operation, which limits the application and
operational functionality and efficiency of the network. In both cases, the regularity of surfacing
is seen as a balance around the needs of the mission and the disruption problems in the
acoustic network when vehicles need to surface. This will impact on the swarm synchronisation
and vehicle energy consumption.
A number of solutions have been proposed for the design of time scheduled medium access
control communication protocols for densely deployed mobile multi-vehicle underwater networks
[36, 87, 113, 115, 123]. Maurelli [87], Schill [115] and Stojanovic [123] have specifically focused
on swarm communication, where intelligent, multi-vehicle AUVs are operating at close ranges
and in quite dense decentralised deployments and thus have the need for short-range
communication using broadcast signalling that provides continuous exchange of information.
Stojanovic [123] discussed the use of a simple TDMA approach to exchange inter-vehicle
distances to build a localisation map in all vehicles which gives several implicit advantages
including finding optimum routing in multi-hop networks and known transmission power
requirements to provide full connectivity. Schill [115] made some modifications to improve the
underwater channel utilisation by changes to time-slot length and slot schedule by using a
reservation based TDMA in a similar way to Kredo [80]. Salva-Garau [113] and Diamant [36] did
not investigate multi-vehicle networks in particular but investigated TDMA based solutions to
short-range high traffic broadcast communication. Salva-Garau [113] used a hybrid
CDMA/TDMA protocol for a multi-cluster network and Diamant [36] investigated the scheduling
of multiple TDMA frames based on spatial reuse through separation of nodes into regional
areas.
The aim of Stojanovic [123] work, which was a very early paper in underwater multi-vehicle
communication, was for mapping the ocean bottom and required explicit communication
between all vehicles in the swarm to build a map of an area by using collaborative navigation,
which uses navigational and sensor information from other vehicles to improve its own location
estimate and incorporates collaborative mapping which uses the exchange of localisation
information of all vehicles to create a more accurate and detailed map of the area being
surveyed. It has been recognised that mapping can be performed faster and more accurately
with multiple vehicles working collaboratively, however with the limitations of underwater
operation the analysis was limited to a swarm of 5 vehicles. A matrix of inter-vehicle distances
and other relevant data, such as Doppler Frequency Shift, relative speed and location, is
maintained in each vehicle and distributed to other vehicles in the swarm using a standard
TDMA based scheduling communication approach. In each frame, each vehicle has an
73
assigned time-slot that it uses to broadcast its up-to-date matrix, whereupon each of the other
vehicles listen and receive the data to update its own matrix. Each time-slot ends with a guard-
time that is set based on maximum propagation delay to ensure that all vehicles received the
data packets without collision. Fundamentally, the protocol proposed is a conventional TDMA
structure, however with much larger guard times and showed that the size of a swarm would be
limited using a TDMA structure.
Schill [115] developed a Distributed Dynamical Omnicast Routing (DDOR) algorithm for a large
number of mobile agents using a RF physical layer. The DDOR protocol incorporates a cross-
layer functionality and combines the networking and medium access control layers. The MAC
layer used a reservation based TDMA, such that during a vehicles assigned timeslot, it can
send firstly a longer message that includes its local schedule and the payload data and then
secondly a very short 2-byte request slot for securing an upcoming time slot. The schedule
provides a means of piggybacking localisation and schedule information from other vehicles,
which is a good feature of this protocol, however the reservation approach does seem
unnecessary in terms of the continuous traffic requirements of the network. A major drawback
occurs for networks with high connectivity which cause longer schedules that create higher ETE
packet delays as well as wasting large parts of the transmitted message portion of a timeslot
and as a consequence having a negative impact on the protocol performance. An improved
algorithm, Pruned Distributed Omnicast Routing (PDOR) [115] was implemented which includes
a determination of the received signal strength so as to resolve packet collisions at the receiver.
It also used signal strength to prioritise the allocation of a request in another nodes requested
slot based on the slot with the lowest locally measured signal strength so as to allow concurrent
transmissions as a technique to implement spatial reuse. The schedule length was found to
reduce from 64 slots to 16 slots in a 40-node network, giving a smaller frame size in PDOR
compared to DDOR and a shorter ETE packet delay. Thus, PDOR modifies both the TDMA
frame length and slot allocation to improve the TDMA performance.
4.3.2 Token Polling Protocols for Swarming AUVs
An early example of a polling protocol used in a small non-mobile network was the ACME
project (Acoustic Communication network for Monitoring of Environment in coastal areas)
sponsored by the European Union [2] and developed the ACMENet protocol which is a polling
based TDMA algorithm for a master-slave topology. This research demonstrated that having a
master controller provided the opportunity to increase the efficiency of the TDMA protocol by
using a polling request process to optimise the use of time slots. The controller was also used in
a network management function as a rule based decision maker to improve efficiency of power
consumption within the node and was able to select the optimum modulation type as a global
parameter of the system. Included as part of this work were extensive sea trials that highlighted
the problematic nature of the underwater environment and the detrimental effects on the
performance of the protocol. High packet loss ratios and link outages were recorded due to the
high acoustic noise generated by shipping traffic. Thus, the robustness of the protocol and error
correction techniques are important to consider.
74
The token-based medium access control protocol (TMAC) was developed for a fleet of AUVs
[42]. It uses a virtual token that is not passed between vehicles but maintained within each
vehicle based on messages transmitted and received. The algorithm maintains a fixed TDMA
cycle time, which for the 5 AUVs used in this research project [8] is a fixed 30 s. The virtual
token is used as a mechanism to trigger vehicles to send in their time slot, primarily to avoid the
need for the TDMA reliance on time synchronisation between vehicles. The token in each
vehicle is updated when it receives or transmits a message and includes either: the expected
vehicle ID and expected time that the next message should arrive from a transmitting vehicle; or
the time to send its message if the vehicle ID matches the token identification number in its
memory. If messages are dropped, token updates are done on a time-out basis, so that the
cycle is not broken. Cycle time and packet ETE delay were not dealt with as large slot times
were used.
A new proposed polling mechanism (UW-Polling) based on roll-call polling was developed for
data uploading from an AUV to a static underwater network and was compared to a random
access approach [45]. Throughput was shown to give a slightly improved performance across a
range of packet generation rates and thereby confirmed that polling is an effective approach to
achieve data uploading in this underwater scenario.
4.4 Challenges and Opportunities using TDMA and Polling Algorithms 4.4.1 Time Synchronisation
Time synchronisation is a significant disadvantage when working with TDMA protocols and this
is particularly problematic underwater where GPS signals are not available. To solve this, two
fundamental different approaches have been developed to deal with time synchronisation when
required by communication protocols and these are either by signal processing or as part of the
protocol design itself.
Syed [129] demonstrated a software-based signal processing time synchronisation algorithm in
a high latency acoustic network which effectively synchronises clock offset and skew in two
phases, suggesting reasonably accurate synchronisation is possible underwater if the
propagation delay is predictable and static for short periods of time. Unfortunately, in swarm
operations this condition is unlikely due to mobility required of vehicles.
Alternatively, there are protocol approaches such as RIPT (Receiver-Initiated Reservation-
Based Protocol) [25], which is a MAC layer protocol for a peer-to-peer decentralized ad-hoc
acoustic network that uses receiver reservation knowledge to build a transmission schedule
which avoids time synchronization as it only requires relative time knowledge. Schill [115] and
Yackoski [144] have both incorporated within their MAC designs a time synchronisation
mechanism which uses the vehicle’s clock timestamp within the navigational data packet
transmissions and then, like RIPT has done, used it to determine collision-free schedules.
A token polling protocol can also provide an alternative approach so that time synchronisation
between vehicles is not necessary by using a token as a trigger instead of a vehicle clock [43].
75
4.4.2 Guard Time
In underwater networks, guard times can be significantly large and therefore very inefficient,
particularly in terms of channel utilisation. Guard times are used in scheduled protocols to allow
for ‘space’ due to an uncertainty in the system or environment. In terms of the use of TDMA in
underwater protocols this ‘space’ or guard period that needs to be added to each fixed time-slot
is due to the long and variable propagation delays and is often calculated by the maximum
propagation delay in the system. In a mobile environment it might be prudent to add even more
time due to the potential change in this maximum propagation delay. For example, if the
network of Figure 4.3 is used, where the range between ID1 and ID2 is 50 m and ID2 to ID3 and
ID3 to ID4 is 150 m, then the maximum propagation delay would be calculated on
approximately 350 m should they be in a straight line and therefore the guard time would be
0.23 s. With the 4 nodes requiring a minimum of 4 time slots, this would add almost an
additional 1 s to each cycle.
Stojanovic [123] recognised the possibility of reducing guard times if intended users were
considered at shorter distances, however as the possibility of packet collision would increase
the need for multiple access interference tolerant receivers was discussed. Yackoski [144] took
quite a different approach to varying guard times in an underwater protocol by designing the
protocol so that guard times were not needed in each time slot but only twice each cycle and
this was shown to reduce the overall packet delay and increased channel utilisation.
Ahn [3] analysed the use of guard times underwater using the ALOHA protocol and showed that
guard times were more influential when transmission time was greater than propagation delay
and that fixed guard times lead to suboptimal throughput. In short-range operations
transmission times can be greater than the propagation delay, and with mobility the required
guard time is dynamic, which explains why guard times are of significant interest in multi-vehicle
swarm TDMA protocols and need careful consideration.
Thus when using TDMA protocol with high traffic requirements and large numbers of vehicles
as expected in a swarm network, guard times are very wasteful of the channel usage. The
advantage that a token polling protocol has is the use of the token as the trigger to transmit or
receive and therefore can avoid the need for guard times.
4.4.3 Scalability
Time-based protocols have limitations in terms of their ability to be scalable. Changing a TDMA
cycle from working with 3 vehicles to 30 vehicles makes a single cycle period extremely long.
There are similar limitations when working with polling protocols whether a roll-call or hub
polling approach is used, as each vehicle still needs to wait its turn to access the medium.
Stojanovic [123] limited the number of vehicles in a network using TDMA MAC layer to 5 but
discussed the provision for allowing vehicles to enter and leave the network. The proposal was
to have a 60-frame period every 10 minutes that allowed new vehicles wanting to enter the
network to speak or if a vehicle left the area during the past 10 minutes its absence would be
noticed and adjustments during this transition time.
76
The study by Salva-Garau [113] investigated the scalability of mobile AUVs using varying
coverage areas and number of vehicles. They used a cluster approach where adjacent vehicles
were put into a cluster with inter-cluster communication based on TDMA while the cluster-to-
cluster communication used CDMA. The time slot allocation was based on lowest identifier
clustering algorithm (LIDCA) to determine user slot schedule within each frame and the first slot
was determined to be the cluster head (CH). The last slot in the frame was for network
maintenance where allocation of vehicles to clusters is determined. This allows movement in
and out of a cluster, which permitted significant flexibility that is beneficial in a fully mobile
environment. Thus this maintenance time occurs much more frequently than in Stojanovic’s
proposal. It also enforces a limitation of equal numbers of user slots in each frame, so that the
maintenance will occur at the same time throughout the network. So, if there are fewer nodes in
some clusters then there will be time-slot gaps in those cluster frames, which will substantially
reduce throughput. Also having smaller cluster sizes and therefore fewer TDMA time-slots per
frame, the ratio of the overheads will increase, as the management slot is a fixed length, and
transmission resources is reduced. Salva-Garau also examined the number of nodes per cluster
to improve network connectivity, however this was influenced more by the operation of the inter-
cluster CDMA protocol. Here they examined the use of a 7-code hexagonal reuse pattern of
clusters (analogous to spatial frequency reuse in cellular networks) and connectivity improved
where there were larger distances between clusters.
For time-based protocols the ability to create very large-scale networks will be influenced by the
reuse patterns and the ability to organise multiple scheduled transmissions at the same time
across a network without packet collisions.
4.4.4 Time-Slot Scheduling
Diamant [36] in a similar approach to Salva-Garau [113] investigated using spatial reuse of
TDMA combined with CDMA by using the DSSS-based signalling in underwater modems to
enable CDMA. The focus of this work is on how to schedule nodes to transmit at the same time
when they are in different parts of the network. Two potential packet collision conflicts were
considered: a primary conflict is where two adjacent nodes are scheduled at the same time and
this conflict is not allowed. A secondary conflict is where a third node receives packets from 2
other nodes in the network at the same time and this can be allowed unless the signal-to-
interference-and-noise-power ratio (SINR) drops below a certain threshold, and thus equivalent
to the well-known near-far problem of single-channel receivers.
Several other researchers, including Hsu’s [61], Kredo [80] and Kleunen [79], have focused on a
graph-theory optimisation approach to allocation of time slots across a network. Kredo
developed STUMP (a staggered TDMA underwater MAC protocol) that uses hop-distance from
the sink with 4 possible packet collision-conflicts at a node, which include Tx-Tx, Tx-Rx, Rx-Rx
and Rx-interference, to determine the constraints needed for the optimisation. Kleunen follows a
similar approach to Kredo but simplifies the 4 possible collision-conflicts at a node by defining
an order of transmission, which effectively eliminates the Tx-Tx conflict and reduces the
complexity around the Rx-interference conflict.
77
Kleunen like Hsu’s ST-MAC protocol lifted the TDMA restriction that requires that scheduled
transmissions need to wait for the start of defined slot times. Kleunen showed that scheduling
within a TDMA or slotted system was less efficient in throughput and thus most likely less
efficient in ETE delay although this was not evaluated, in a centralised topology. They also
concluded that differences in slotted and unslotted scheduling methods become smaller when
the ratio of transmission time to propagation delay becomes large. Hsu summarised that spatial
uncertainty has a great impact on the performance of underwater MAC protocols, showing up to
97% higher throughputs using Spatial-Temporal Conflict Graphs than conventional underwater
MAC’s such as T-Lohi [129].
Therefore, to improve channel utilisation by: allowing concurrent transmissions; reducing packet
collisions at the receivers; and reducing the guard times necessary for this, the schedule of
access needs to be designed intelligently around the spatial-temporal environment experienced
when using acoustic communication. In addition, when considering USSN applications the
question is also how can spatial-temporal diversity be developed for decentralised networks.
4.4.5 Spatial-Temporal Diversity
The shorter range between vehicles in a swarm topology, considered in this work, means that
propagation delays will be smaller than for the more typical longer-range underwater
applications. In fact, the transmission time for packets underwater in short range scenarios will
be in the same order of magnitude as the propagation delay, which creates a unique spatial-
temporal environment for underwater swarm communication, and is far different from what is
experienced in a terrestrial RF setting.
Syed [129] & Yackoski [144] have illustrated this in the more general case by showing that both
the times of transmission and range to receiver need to be considered in underwater
environments to avoid packet collisions, when compared to traditional analysis where only the
time of transmission is considered. That is, in RF, the propagation delay is considered
negligible, which means that packets that start transmissions at the same time will collide at the
receiver and thus RF protocols are generally designed for ‘exclusive’ channel access to avoid
collisions.
An exclusive channel access condition does not need to be maintained in underwater
environments as will be explained using Figure 4.3. Figure 4.3 represents a four-vehicle
network, ID1 to ID4, and shows one moment in time, where ID3’s packet was transmitted earlier
than ID1 and ID4 packets. The thickness of the band (red, green and blue) represents the
transmission time of the packets, and each are equal thickness meaning that the three packets
are of equal length. Despite ID1 & ID3’s packets starting transmission at different times, their
packets will collide at ID2, while the concurrent transmissions of the ID1 and ID4 packets will not
collide at ID3 or ID2. That is, it can be seen in the Figure that ID1’s packet has already reached
ID2 and will have completed reception well before ID4’s packet is likely to arrive. Exclusive
channel access based on transmission time of data becomes an ineffective way to avoid
collisions, as large guard times need to be incorporated to take into account the slow signal
78
propagation between vehicles in the network. Therefore, non-exclusive access can occur due to
the space-time diversity, which allows more than one transmission-reception activity in the
channel at the same time.
To exploit this spatial-temporal diversity, the vehicles relative position and therefore propagation
times between other vehicles together with the packet size and system data rate, which
determines transmission time, is required. The ratio of propagation to transmission time, β,
demonstrates more precisely the relationships between range, packet size and data rate for a
point-to-point link.
𝛽 = 𝑡𝑝𝑡𝑡
= 𝑡𝑝 𝑅𝐿
= 𝑟𝑎𝑛𝑔𝑒∗ 𝑅𝐶∗ 𝐿
(4.1)
where tt is the transmission time 𝑡𝑡 = 𝐿𝑅 (s), R is the channel data rate (bps), and L is the length
of packet (bits), and tp is the propagation delay 𝑡𝑝 = 𝑟𝑎𝑛𝑔𝑒𝐶
(s), C is the speed of acoustic
propagation underwater which varies around 1500 m/s.
There are three potential scenarios:
β < 1; when the transmitter will not have finished transmitting the packet when the receiver will
have begun receiving it. Here the vehicles will be very close together
Figure 4.3: Spatial-Temporal Diversity
79
• and/or packet length is long so that there is an overlap time of transmission and
reception of the same packet. In the RF case β << 1,
• β = 1; when the propagation time equals the transmission time,
• β > 1; when the transmitter will have finished sending a packet before the distant
receiver starts to receive it. In this case there is a ‘void’ period between sending and
receiving the same packet. In typical longer range underwater systems, β >> 1.
Figure 4.4 illustrates the affect of each scenario on a simple time slot protocol. The overall
cycle for exchange of messages between the two nodes increases as β increases. As can
be seen, in all cases there is a ‘transmission gap’ in each nodes time slots and as β
increases so does the gap size. In the RF case where β << 1, that is the propagation delay
(tp) between a sender and receiver is several magnitudes smaller than the transmission
delay (tt) of the data (tt >> tp), the transmission gap is considered negligible which is why an
exclusive access channel is generally assumed. This allows the channel utilisation ( 𝑡𝑡𝑡𝑡+𝑡𝑝
) to
approach 100%, where tp incorporates the guard time.
This is not the case for underwater communications where β ≥ 1 and the transmission gaps
are greater than the transmission time of the data packet themselves. Here the assumption
of an exclusive channel requires the incorporation of very large guard times if collisions are
to be avoided, which is why exploitation of the gap is necessary to improve channel
Node 1 n1t n2a β << 1
Node 2 n1a n2t
Node 1 n1t n2a β < 1
Node 2 n1a n2t
Node 1 n1t n2a β = 1
Node 2 n1a n2t
Node 1 n1t n2a β > 1
Node 2 n1a n2t
Node 1 n1t n2a β >> 1
Node 2 n1a n2t
tt + tp
Figure 4.4: One Data Exchange Cycle between 2 Nodes for Different β
80
utilisation. Typical longer range acoustic communication underwater operates with β >> 1,
however for swarm operations the network will be operating around the β = 1 region where
guard times can be reduced but still need to be incorporated if an exclusive channel assumption
is made.
With time-critical data communications, which is a major issue in swarm operations, utilising the
transmission gaps will increase the channel utilisation efficiency and decrease the time between
receiving exchange updates and thus should improve throughput and latency performance. To
do this, non-exclusive transmission needs to be implemented which can be done with careful
scheduling.
4.4.6 Application of Spatial-Temporal Diversity
In a contention based MAC protocol, T-Lohi, Syed [129] undertook to exploit the spatial-
temporal diversity underwater by using the idea of contender counting once detection had
happened during the contention round to allowing concurrent transmissions in the reservation
period if packet collisions could be avoided. Yackoski [144] and Schill [114] also proposed
variations of this idea to gain spatial reuse of time slots across the network by modifying a
conventional TDMA protocols to integrate a contention period during each cycle. Both added an
adaptive reservation period to the TDMA by splitting up time intervals to have a shorter
reservation period and a longer transmission period and they both use a database of neighbour
node information that includes local schedules of ‘bad times’ that collision will occur for each
node. The main difference is that Schill has split each time slot into a reservation and
transmission portion whereas Yackoski splits a whole cycle into the two portions; an
Establishment period, which all nodes compete to reserve a time to send data and a
Experimental portion where the messages get transmitted based on a relative time system.
Yackoski’s protocol is designed for peer-to-peer operation rather than a broadcast mode and
works better for stationary nodes, since before nodes can request a slot they need to know both
the global and local ‘bad times’ to determine a collision-free time to fit the next packet which is
difficult to adapt quickly enough to for mobile networks. Yackoski implemented a time
synchronisation mechanism through the exchange of each nodes clock times to develop a
method that relies only on relative times. In addition, as mentioned earlier in Section 4.4.2,
Yackoski’s protocol only requires two guard periods a cycle, which is between the establishment
and experimental portions, rather than every timeslot, which therefore reduces the time
overheads. However, for a mobile network with continuous traffic the reservation approach
becomes superfluous and impacts on the efficiency of being able to exploit all the unused times.
Diamant [35] also utilises a neighbourhood database called a Connection List (CL), which
identifies the 1-hop distant ‘receiving nodes’ and the nodes that are out of transmission range or
2-hop distant nodes that are called ‘joining nodes’. As in a conventional TDMA approach, Node
A is assigned timeslot t=A, however to improve the Successful Packet Transmission Rate
(similar to throughput), spatial reuse is implemented by nodes using the CL to adaptively
identify available nodes that can transmit during the same timeslot without collision which can
only be in the joining nodes category. To eliminate conflict among the joining nodes, only the
81
even-hop distant joining nodes can be candidates for transmitting at the same time while all the
odd-hop distant nodes must remain receiving nodes during that timeslot. As discussed in
Section 4.4.4, Diamant [36] also uses a CDMA physical layer to allow 2-hop conflicts to be
avoided in a similar way to Salva-Gaura [113] in their TDMA/CDMA MAC layer multi-cluster
protocol.
4.4.7 Summary
The need for spatial-temporal scheduling arises in underwater acoustic communication
predominately to improve channel utilisation, which addresses not only throughput but also ETE
delay performance. In particular, the use of space diversity relates to range between nodes (or
tp), where two nodes can send at the same time without collision at the receiver, either from
propagation delay or from signal characteristics at the receiver and with the use of temporal
diversity that relates to variations in tt and data rates where the size of the blocks of time
occupied by a message will impact on the probability of collision. Working with these two
phenomena means that instead of considering the long propagation delay, due to speed of the
acoustic signal, as a drawback it will be investigated as an advantage to schedule multiple
accesses in the time while avoiding collisions and using signal characteristics to create time
reuse patterns.
This work will focus on transmission resource reuse patterns using single-hop single-cluster
swarming topologies to improve throughput and ETE delay performance which is different to the
reuse patterns developed across multi-cluster networks as seen in Salva-Garua [113] and
Diamant [36]. Also, instead of developing a hybrid reservation based TDMA protocol (Schill
[115] and Yackoski [144]), the use of token polling mechanisms will be explored.
4.5 Conclusion Research efforts into Underwater MAC protocols for Autonomous Swarms of AUVs has been
very limited, yet as AUV technology advances, the need for cooperative communication within a
swarm of AUVs will be essential to allow them to operate in close proximity of each other and
utilise the emergent properties and behaviours of a swarm. The communication traffic required
in a closely operating swarm of AUVs, is for continuous traffic with quick local and global
information dissemination in a highly interconnected network. This, as we have discussed, lends
itself to a scheduled medium access control approach, as these protocols are designed to avoid
packet collisions and thus the wastage of transmission resources that they can generate. The
disadvantage, however, is the fixed and inflexible nature of the schedules in terms of both slot
time and slot allocation.
More recently, the understanding and incorporation of the unique spatial-temporal qualities of
an underwater acoustic channel have been investigated. These protocols attempt to leverage
knowledge of node position diversity and transmission timings due to the slow speed of sound
propagation underwater to make improvements in channel utilization. In particular, they
recognise the dependence of both the time of transmission and the range to a receiver to allow
for and avoid collisions.
82
The goal of this work is the development of a new advanced medium access technique for
efficient underwater swarm communication to enable both Mission Time Critical and Mission
Non-Time Critical applications. Specifically, it is to develop a Medium Access Control (MAC)
algorithm for a robust mobile decentralised single-hop swarm sensor network that will minimise
latency, maximise usable bandwidth and allow the dissemination of time-sensitive data
throughout the swarm in a timely fashion. The option of using concurrent transmissions appears
possible and beneficial and this will be particularly focused on in the new proposed protocols. In
particular, mechanisms to advance flexibility in slot time and slot allocation will be incorporated
to increase the availability of transmission resources to each vehicle for throughput and packet
latency performance improvements.
83
Chapter 5 Introduction and Analysis of Two New MAC protocols for Underwater Swarm Sensor Network (USSN) Applications
5.1 Introduction Two new communication MAC layer protocols are introduced in this chapter based on the
different network architectures required by the two application categories defined in Chapter 2
for Underwater Swarm Sensor Networks. Both new protocols incorporate the use of ‘non-
exclusive channel access’ a method to handle and exploit the unique channel characteristics
experienced underwater, that include in particular, low bandwidths and long propagation delays
and are designed to operate in a single channel broadcast acoustic environment. The protocols
are based on time divisions, however relying on the use of a token to trigger new transmissions
as opposed to a clock in the more traditional TDMA MAC layer protocol.
The first application area is the Non-Time Critical Mission such as surveying or monitoring that
are best suited to a vehicle configuration that maintains a structured formation pattern that can
sweep an area in a structured way. A Bus Topology with a coordinator vehicle is proposed so
that an area could be covered systematically with a preprogramed survey pattern. The
coordinator vehicle is incorporated to provide a network view of the operations to improve
efficiency of application requirements should some modification of the preprogramed pattern be
beneficial as well as for co-ordination of the communication requirements. The Adaptive Token
Polling MAC (ATP-MAC) which uses a token polling ring approach allows the slot size to be
adaptive based on the transmission time and propagation time between consecutive vehicles.
The ATP-MAC was first developed as a centralised communication architecture [15] but can
also be used in a decentralised hierarchical structure to reflect the more traditional definition of
a swarm described in Section 2.4. It will be the later approach of a decentralised structure that
will be presented and analysed in this chapter.
The Time Critical Mission, such as for search and rescue applications, requires that the swarm
dynamically responds to local and up-to-date information that is collected by each vehicle in the
swarm so that together they can redirect the swarm to find an object or complete a task such as
finding a hydrothermal vent. That is, the swarm can move to the items of interest directly rather
than via a planned pattern. This involves swarm intelligence where the swarm vehicles
maintain a cluster in which individual vehicles move in a random fashion based on a regular
trajectory calculation that is done by each individual vehicle from localisation information just
collected from its neighbours. Thus the MAC layer protocol is critical as explicit communication
is required between vehicles. A new MAC layer protocol has been proposed for this swarming
network, the Adaptive Space Time – TDMA (AST-TDMA) [17], that uses a TDMA approach and
also a token in a similar fashion to ATP-MAC to facilitate an adaptive slot size that more
84
efficiently accommodates the long propagation delays that occur underwater. AST-TDMA,
however is a fully distributed decentralised MAC layer protocol that does not rely on a
coordinator vehicle and reflects the requirements of a traditional defined swarming network.
The Bus Topology can be a linear 1-Dimensional (1-D) structure as a straight line or 2-
Dimentional (2-D) wedge shape shown in Figure 5.1 and is designed to maintain this shape
while in operation, whereas the Cluster Topology is designed for 2 and 3 Dimensional (2-D and
3-D) operations. This chapter, however, will concentrate on the description and abstract
analysis of the two protocols in 1-D and their comparison with a conventional TDMA protocol in
both topologies. Chapter 6 will develop and simulate in 2-D formation a swarm of AUVs using
the AST-TDMA protocol and explore the possibility of separating time-sensitive and non-time-
sensitive traffic data sources. The introduction of a realistic acoustic communication channel
based on the work in Chapter 3 will also be used to evaluate the protocols further to establish
their performance credibility.
This chapter will begin with a discussion of the different deployment strategies required for the
application areas previously considered. A description of the two new protocol structures and
operations are explained together with the different packet types required and their structures.
The performance criteria and new metric, NCCP (Neighbourhood Communication Cycle
Period), are discussed together with the Quality of Service (QoS) boundaries. A queuing model
is used to provide an analytical analysis of the communication protocols and to study the
suitability of these protocols to their applications. The work in this Chapter will be based on ideal
conditions, with swarms operating in a stable structure, meaning that the range between
vehicles will not change in either a 1-D/2-D formation. This will provide the base line values for
comparison with the evaluation of the protocols in ideal and non-ideal simulations presented in
Chapters 6.
5.2 Application Deployment Strategies The two categories of application areas that have been investigated are a Non-Time Critical
Mission, used for formation surveying, and a Time Critical Mission, used for dynamic searching,
as discussed in Chapter 2 (Section 2.5). These two applications require a significantly different
deployment approach and data traffic requirement, as is summarised in Table 5.1. This section
will describe each application area in more detail.
The terms ‘time-sensitive’ and ‘time-critical’ are used to differentiate the impact that the
navigational data exchange timing has in each scenario. Time-critical is used as an application
layer QoS parameter for the Time Critical Missions as the timing of the exchange is required to
ensure avoidance of vehicle collisions. In the Non-Time Critical Missions a preloaded formation
control pattern provides the avoidance of vehicle collisions, however, due to unforeseen
disturbances, exchange of navigational data is required to maintain relative positioning and
therefore is refer to as time-sensitive. This is a MAC layer QoS issue on how quickly packets
can be sent.
85
Table 5.1: Application Specific Deployment and Communication Requirement Overview
Application Non-Time Critical Mission Time Critical Mission
Formation Patterned (lawn or lane etc.) Swarm
Topology Bus Cluster
Navigational Data
Exchange Requirements Time-Sensitive / Constant
Range dependent. Time-Critical
to Time-Sensitive/ Constant
Payload Data
Exchange Requirements Application-Dependent /
Generally constant
Can be used in Navigational
Data. Application-Dependent /
Constant or Poisson
MAC Protocol ATP-MAC AST-TDMA
5.2.1 Non-Time Critical Mission Deployment
Vehicle deployment for surveying and monitoring applications necessitates a structured and
stable pattern of motion that offers a consistent and steady sweep of an area. There are
numerous variations to the arrangements of vehicles in a Bus Topology such as simply using a
single line or V-shape formation as seen in Figure 5.1.
Figure 5.1: Bus Topology for a Non-time Critical Mission using an Underwater Swarm Sensor Network
Surface
Sea Floor
Swarm
Vehicles
Coordinator Vehicle
86
These swarm networks can operate with heterogeneous or homogenous groups of AUVs which
will not affect the design of the MAC layer protocol. A homogeneous group of AUVs fit the more
typical swarm definition and so with this deployment arrangement there will need to be one of
the vehicles nominated as an initial coordinator vehicle to initiate the first poll. The network then
can initiate a mechanism to share the role around the swarm using a rotation arrangement. This
also has the advantage of more even distribution of energy consumption among the vehicles
[103, 115]. Alternatively, the swarm could be made up of a heterogeneous group of vehicles
where a more sophisticated vehicle can sustain taking on the coordinator role for the whole
mission [116, 146].
5.2.1.1 Non-Time Critical Mission Data Traffic
Due to the predetermined path and formation pattern required by this application where a pre-
set route for each vehicle is loaded prior to operation, the exchange of navigational data is
required by these closely operating vehicles to maintain relative positioning and swarm
synchronisation, which will be discussed in more detail in Section 5.7 as a essential Quality of
Service (QoS) criteria. The exchange of navigational data is therefore time-sensitive but not
time-critical as the pattern formation provides the overall trajectory control and the navigational
data traffic is exchanged to ensure that vehicles do not collide and can provide dynamic relative
adjustments to individual vehicles trajectory.
The payload data for these applications is generally collected from each vehicle at the end of
the mission for post-mission analysis. However, payload data may be useful to exchange for
redundancy of mission data, in case of vehicle failure or loss, or to support dynamic surveying
where for example it might be determined that a more concentrated survey would be more
beneficial. This could be done during a mission by readjusting the formation pattern and
coverage area of the swarm.
Figure 5.2: Cluster Topology for a Time Critical Mission using Underwater Swarm Sensor Network
Surface
Sea Floor
Swarm
Vehicles
87
5.2.2 Time Critical Mission Deployment
The deployment of vehicles in Time Critical Mission applications are influenced by the real-time
payload data collected and the location of neighbouring vehicles as defined by the biologically
inspired formation control algorithms discussed in Chapter 2.6.3. This creates a random pattern
of formation as vehicles manoeuvre in a swarm-like fashion and is referred to in this work as a
Cluster Topology.
The Cluster Topology, illustrated in Figure 5.2, reflects the standard definition of a swarm, which
can be classified as having a distributed decentralised architecture. In these networks it is more
common that the swarm is made up of a homogenous group of vehicles where all are identical,
however this is not strictly important from a communications capability point of view. In this work
the assumption will be that each vehicle will have identical and independent capability in terms
of communication mechanism.
5.2.2.1 Time Critical Mission Data Traffic
Time Critical Mission applications include searching and finding objects or tasks where using
the knowledge of the swarm (swarm intelligence) can be utilised to increase the speed of the
search. These networks require a swarm formation control algorithm for the determination of the
dynamic trajectory of each of the vehicles and the overall direction of the swarm based on
fulfilling a mission. The swarm formation control algorithm will define the data traffic required in
each application and generally will include the localisation information from neighbouring
vehicles to avoid vehicle collisions and payload data that will support the mission outcomes by
directing the overall direction of the swarm. The option to exchange payload data more directly
may or may not be necessary depending on application and data redundancy requirements.
The localisation data traffic that is required to be exchanged among all neighbouring vehicles in
this application is time-critical for swarm synchronisation, which is a QoS requirement as this
data is required to avoid vehicle collisions and to maintain formation in a swarm like-fashion.
5.3 Adaptive Token Polling (ATP-MAC) Protocol Description The Adaptive Token Polling MAC Protocol has been designed as a MAC layer protocol for a 1-
D Bus Topology in an underwater swarm network for the Non-Time Critical Applications. These
applications benefit from a coordinator vehicle that can provide an overview of the structured
formation and thus can utilise a hierarchical architecture. The ATP-MAC protocol is a
decentralised hierarchical network that utilises a polling mechanism that enables an adaptive
timing structure to take into account the long propagation delays that are experienced in an
underwater acoustic communication channel.
The ATP-MAC has adapted a centralised hub-polling algorithm [21], however with a major
difference in the usage of the poll packet which for the ATP-MAC protocol incorporates the
management and data information for the whole swarm and will therefore be referred to as the
Command Packet for the swarm. Similar to a polling packet, the Command Packet is sent out
by the coordinator vehicle to signify the start of the cycle and to provide a token or permission
88
for the next vehicle to access the medium but it also provides the control of swarm direction by
providing trajectory information to all the swarm vehicles and therefore contains a large amount
of data which is substantially larger than a typical poll packet. ATP-MAC thus exploits the use of
a coordinator vehicle, which has an overview of the network, to provide control of the swarm
direction and management of the communication cycle through its Command Packet.
The more common roll-call polling [21, 11] is not used here due to the high latency channel.
This is because in roll-call polling the poll is designed simply to give permission to one vehicle at
a time to transmit its packet and this therefore requires two propagation delays for each data
packet sent; which is the transmission of the poll packet to the vehicle and then the data packet
in return back to the coordinator vehicle. To avoid this a token polling method is used which
uses the coordinator vehicle’s Command Packet to manage the swarm and begin the cycle by
including the first token. This avoids the overhead of twice the propagation delay, which is
particularly significant in the high latency environment experienced underwater. The proposed
Command Packet from the coordinator vehicle can also include management of the
communication cycle by determining on a regular basis an optimum schedule of transmissions
and then initiating the new schedule with the token or changing the modulation type and power
levels to increase communication efficiency as was studied in ACMNET [2].
The proposed protocol is adaptive in terms of slot length by using a series of tokens that pass
the permission to access to the medium from sequenced vehicle to the next sequenced
vehicles. When a swarm vehicle receives the token, this triggers it (and no other vehicle) to
transmit its data packet back to the coordinator vehicle and thus a series of slots are allocated
to the sequenced vehicles. These slots will vary in length from slot-to-slot and cycle-to-cycle as
they are determined by the propagation time between the sequenced vehicles. The token
effectively provides a means to avoid accurate time synchronisation between vehicles as the
token provides the trigger instead of the use of a clock to trigger the beginning of each slot. This
also improves transmission efficiency compared with a TDMA based protocol, as slots do not
need to be held open if there is no data to send.
Another major advantage of using a token is that guard times are no longer needed in the ATP-
MAC protocol, as the token is the confirmation that the transmission is completed. Thus passing
a token avoids packet collisions, which lead to very long delays with retransmission
requirements and low throughput. The ATP-MAC has incorporated a processing time for a
packet’s transmission and reception, which provides a small gap that acts in addition like a very
short guard time. The TDMA protocol however incorporates large guard times that are required
with long propagation delay environments to avoid slot overlap and timing errors in packet
reception with the fixed time slot arrangement.
A coordinator vehicle’s overview of the network is also beneficial in terms of operational
efficiencies. Having the coordinator vehicle maintain knowledge of the location of each of the
swarm vehicles provides a means of ensuring that vehicles maintain their position in the Bus
Topology relative to the other vehicles and allows the coordinator vehicle to build a map of the
89
survey area on which the other ‘sensed’ data can be overlayed. It is usually important that the
payload data be collected in association with the vehicle’s position as knowing the sensed data
location allows revisiting that location should a ‘hole’ in the data be found. A vehicle failing or
being lost in a swarm can be dealt with autonomously when there is a coordinator vehicle as it
can control revised localisation of swarm vehicles by reassignment of vehicles’ positions within
the Bus Structure. A major disadvantage of having a coordinator vehicle is the reliance on that
vehicle and the Command Packet that it sends each cycle. Having a rotational coordinator
position may provide some redundancy should this occur.
The Command Packet is broadcast from the coordinator vehicle and indicates the start of a new
cycle. This packet provides the synchronization and management of the swarm. The major
information the Command Packet carries is the directional updates and new trajectory
requirements for each of the swarm vehicles, which has been determined by the coordinator
vehicle based on information obtained from the data returned in the previous cycle. The
Command Packet also includes the first token, which will identify and give permission to the first
swarm vehicle of that cycle to send its data packet back to the coordinator vehicle. Included in
each swarm vehicle’s data packet is a token to identify the next swarm vehicle to send data and
so on. Each swarm vehicle can then either send a data packet if it has data in its queue or if it
has no data packet in its queue then it sends only a token packet to ensure the cycle is not
broken. If it is the last swarm vehicle in sequence, then the token returns to the coordinator
vehicle and the cycle begins again.
The data packets are also broadcast so that all vehicles within the swarm can overhear each
other and thus have some knowledge of the status of the channel as well as receive the tokens
to determine if it is the next swarm vehicle with the right to start transmitting. The token polling
process requires that the coordinator vehicle collects and processes each of the swarm
vehicle’s information; this does not need to be done in each individual vehicle.
5.3.1 ATP-MAC Packet Structures
There are three types of packets required for this architecture: a Command Packet from the
coordinator vehicle, a Data Packet from the swarm vehicles and a Token Packet from the
swarm vehicles if they have no data to send. The data traffic required therefore is that the
coordinator vehicle provides trajectory modification and control to the swarm vehicles while the
swarm vehicles need to reply with their local information so that the coordinator vehicle can
update everyone’s position. This local information will therefore include the swarm vehicles
localisation data and may include payload data.
The size and structure of these packets are illustrated in Table 5.2. The frame structure is
based on the IEEE802.5 Token Ring protocol [83 pg 413]. The token and data frame structure
of IEEE802.5 have been tailored for the requirements of this scenario. It is acknowledged that
these are large packet sizes for typical underwater use and that data and/or packet header
compression techniques should be applied to reduce these sizes.
90
Table 5.2: ATP-MAC and AST-TDMA Packet Structures (bytes)
Field Headers Data Error End InfoType SD AC FC Addresses Command Navigational Payload FCS ED FS Command 1 1 1 12 100 - - 4 1 1 Data 1 1 1 4 0 40 30 4 1 1 Token 1 1 1 4 - - - 4 1 1 where SD is Starting Delimiter, AC is Access Control, FC is Frame Control, FCS is Frame Check Sequence, ED is
Ending Delimiter and FS is Frame Status. Addresses include the Source and Destination Address information. The
additional packet required by the ATP-MAC protocol
Due to the low data rates available underwater and the benefits shown using smaller packet
sizes [125], the aim was to minimise packet size by ensuring low packet overheads and
consolidating the information data portion of the packet. The Token Packet used here is longer
than that used in IEEE802.5, which is just 3 bytes, as the token needs to be a control packet in
its own right. The token packet is required to inform the coordinator vehicle of where it has
come from and that it did not have navigational data to send and the destination vehicle address
to which it will pass the token to give the next vehicle permission to access the channel.
The Command Packet has been designed based on the IEEE802.5 data frame rather than the
token frame and is the largest packet in ATP-MAC due to the need to include management
information for the network. This will include the order in which the token should be passed in
the next cycle and navigational information to each vehicle such as minor adjustments or major
changes to the pre-programmed planned trajectories. The size of the Command Packet has the
potential to vary due to the number of vehicles in the swarm as well as from cycle-to-cycle within
an operating swarm due to the requirements of changes to the swarm vehicles trajectory. In this
work a fixed Command Packet size will be used.
Thus the Command Packet is 121 bytes (968 bits), the Data Packet is 53 bytes (424 bits) with
navigational data only and the Token Packet is 13 bytes (104 bits).
5.3.2 ATP-MAC Cycle Description
A typical single cycle is abstractly illustrated in Figure 5.3 using a four-vehicle network with VL
(coordinator) and three other swarm vehicles (V2 , V3 , V4). The ATP-MAC protocol has two
phases; the command (or polling) phase and the data exchange phase and requires
unidirectional communication. The command phase is the initial slot of the cycle when the
coordinator vehicle has access to the medium. In this slot the coordinator vehicle broadcasts its
Command Packet to the swarm vehicles. The swarm vehicle that is the target of the Command
Packets token, vehicle V2 in Figure 5.3, gets given the permission to access the medium. At this
time the slot finishes at Tx + tL2. The Command Packet continues to propagate and will be
received by the other vehicles in its neighbourhood. The second phase, or data exchange
phase, of the ATP–MAC protocol begins and includes the remaining slots of the cycle. That is,
V2 begins the next slot by broadcasting its data packet and the slot will end when the target of
its token, vehicle V3, has received the packet. Again this packet will continue to propagate and
be received by other vehicles in the swarm including and most importantly be returned to the
91
coordinator vehicle. This continues until the token has been passed to all swarm vehicles and
they have each in turn broadcast their packet that includes the token for the next vehicle in
sequence and their data which is to be returned to the coordinator vehicle.
Thus a cycle is a back-to-back set of transmissions starting with the command packet from the
coordinator vehicle and ending with the coordinator vehicle receiving the final packet from the
last vehicle in sequence. Although broadcast packets are sent so that all vehicles can overhear
packets from each other the timing is based on the sending and receiving of a packet from the
vehicles in sequence only. The cycle will include a Command Packet and a set of packets from
each of the swarm vehicles. The transmission times of the Command Packet, Tcomm, and the
Data and Token Packets, Tdata and Ttoken, are generalised and represented by Tx in Figure 5.3.
Each swarm vehicle transmits either a Token Packet if there is no data to send or a combined
Data Packet that includes a Token. Slot times will also vary if only a Token Packet is send from
a vehicle where there is no data in its queue, as the Token Packet is much smaller than the
Data Packet and therefore the transmission time is much smaller.
* Difference between the protocols is the use of a Command Packet in ATP-MAC. In the AST-TDMA protocol VL would
become V1 and ordinary Swarm Vehicle and would send either a Data or Token Packet similar to the other Swarm
Vehicles
Figure 5.3: ATP-MAC and AST-TDMA Protocol Operation Showing One Full Cycle of Transmission
92
The propagation time between sequenced vehicles (tL2, t23, t34, t4L) will vary with range between
these vehicles, which will be especially true with any vehicle movements, and these times will
have a major impact on determining the cycle time. As each slot time can vary based on packet
size and range to the next sequenced vehicle, so will the cycle time, which will therefore also
vary from cycle-to-cycle. The cycle time is based on propagation times between the sequenced
vehicles as will be discussed further in Section 5.3.3, however due to omni-directional
transmission the packet will continue to propagating to other vehicles in the swarm as shown in
Figure 5.3. This non-exclusive access is allowed because of the space-time diversity in
underwater environments, which is the advantage that this protocol gains and will be explained
further in Section 5.5.
5.3.3 Cycle Time (Tcycle) Analysis
The average protocol cycle time 𝑇𝑐𝑦𝑐𝑙𝑒������� can be determined by adding the time intervals illustrated
in Figure 5.3. Assuming there are V vehicles in the swarm, of which one is the coordinator
vehicle and thus there are V-1 swarm vehicles. The fixed time components in the cycle are the
transmission times of the Command Packet Tcomm that incorporates the first token and of the
Token Packets or token part of the Data Packet from the swarm vehicles ((V-1)Ttoken). As the
Command Packet is required to incorporate information just collected from the previous cycle,
and is the packet that always triggers the beginning of a cycle, a calculation period to create the
Command Packet is included, tcal. An additional processing time is also included for reception
and transmission of all packets. The processing time, however, will depend upon the packet
length and the processing capability of the receiver, which will be estimated at 100 kbps. For the
analytical analysis an average processing time per packet (𝑡𝑝𝑟𝑜𝑐𝑒𝑠𝑠���������) will be used.
The variable portion of the cycle time includes the propagation delays between sequenced
vehicles and the additional transmission time if a Data Packet was sent (𝑁𝑑���� (Tdata - Ttoken)),
where 𝑁𝑑���� is the expected number of data packets sent in a cycle. Td-t will be used, Equation 5.1,
to represent the size of the Data Packet without the Token Tdata - Ttoken, as the size of the token
packet is sent every cycle. A cycle will include the propagation times between each of the
sequenced vehicles (tL2 + t23 + t34 + t4L) where for this analysis a stable structure has been
assumed so that the ranges will be fixed. The range between swarm vehicles is maintained to
be the same and the range to the coordinator vehicle may or may not be equal to this fixed
range. To generalise for this analysis the average propagation time between swarm vehicles is
𝑡𝑆𝑆 ���� and the average propagation time between the coordinator vehicle and a swarm vehicle is
𝑡𝐿𝑆����.
The expected time for one cycle 𝑇𝑐𝑦𝑐𝑙𝑒������� is therefore:
𝑇𝑐𝑦𝑐𝑙𝑒������� = 𝑇𝑐𝑜𝑚𝑚 + 𝑡𝑐𝑎𝑙 + (𝑉 − 1)𝑇𝑡𝑜𝑘𝑒𝑛 + 2𝑉𝑡𝑝𝑟𝑜𝑐𝑒𝑠𝑠 + 𝑁𝑑����𝑇𝑑−𝑡 + 2𝑡𝐿𝑆 ���� + (𝑉 − 2)𝑡𝑆𝑆 ���� (5.1)
93
5.4 Adaptive Space Time - TDMA (AST-TDMA) Protocol Description The Adaptive Space Time – Time Division Multiple Access MAC Protocol (AST-TDMA) is
designed for an operational swarm in a 2-D or 3-D space for the Time Critical Mission
applications. These applications require a protocol where the collective intelligence of a swarm
may be used to search for and find objects. That is, the Cluster Topology of autonomous
vehicles needs to operate in a distributed autonomous random swarm-like formation. Therefore
the aim was to develop a MAC layer protocol that was fully distributed and enabled complete
autonomy and thus could not rely on a coordinator vehicle, as is required by the ATP-MAC
protocol, but instead operate with a group of equivalently simple vehicles in an unstructured
swarm fashion; the classic definition of a swarm (Section 2.4). Removing the coordinator vehicle
means that the polling mechanism could not be implemented, however the advantages of using
a token have been maintained.
This new protocol is most closely based on a TDMA approach where vehicles are allocated a
slot in a cycle rather than the token polling of the ATP-MAC protocol. The AST-TDMA protocol
does not have a command packet but requires that each vehicle determines its own next
trajectory based on information shared as stipulated by the swarm formation control algorithms
discussed in Section 2.4.1. Similar to the ATP–MAC protocol but unlike the TDMA protocol
AST-TDMA uses a token for controlling the access to the communication channel by each
vehicle which means that the slot length is not fixed but will adaptively change, from slot-to-slot
and cycle-to-cycle, due to the changes in range and therefore propagation delay resulting from
vehicles’ position in the swarm and swarm movement. Thus, similarly to ATP-MAC, the AST-
TDMA protocol uses a back-to-back transmission series based on the vehicle’s sequence in the
swarm. That is, with the arrival of a packet at a vehicle, which includes the token identifying it as
the next vehicle to access the channel and therefore permission to transmit its Data Packet in
the next time-slot. This trigger is exactly the same as described for the ATP-MAC protocol and
allows packets to be sent immediately rather than having to wait until the end of a slot time as
required in the TDMA protocol. Therefore, it is only the transmission time and propagation delay
between these two vehicles that determines each slot length. As slot length varies, so will the
cycle time for the exchange of each vehicles’ information. Using a token allows the AST-TDMA
protocol to retain the advantage of not needing guard-times, except for a small processing time
buffer required for the reception and transmission of packets, and the need to provide relative
time synchronisation that are otherwise major disadvantages in time-based protocols such as
TDMA.
The token is thus an important feature of the protocol and is used in a similar way to the token
in the ATP-MAC protocol, except now the token follows a continuous cycle similar to a TDMA
approach. That is, cycles repeating based on a vehicle schedule compared with the use of a
‘poll’ or command management mechanism at the beginning of each cycle in the ATP-MAC
protocol. In fact, the AST-TDMA protocol operates in a similar fashion to a token ring, in the
context that, when a packet arrives at the vehicle, which is to send out the next packet, it acts
94
like a token or trigger to give permission to that vehicle and no others to transmit. The token
thus allows the next vehicle in sequence to access the medium and triggers the beginning and
end of each time slot.
When using the AST-TDMA protocol, like with ATP-MAC protocol, each vehicle will be triggered
to transmit when it receives the token indicating it has permission to use the next slot. In the
simplest of cases the vehicles do not need to have the complete slot sequence and if vehicle ID
is used as the determinant of the sequence, as shown in Figure 5.3, then the vehicle with the
lowest ID will need to have prior knowledge of the number of vehicles within the neighbourhood
or swarm to define the end of the sequence so that the sequence can start again. The vehicle’s
ID will be initially used to determine its sequence position, however adapting to changes in
swarm size due to vehicles being lost or failing needs to be considered. Optimising slot
sequencing to adapt to vehicles changing position within the swarm is out of the scope of this
work, but would be an option to replace the advantage of the coordinator vehicle being able to
do this in the ATP-MAC.
In the fully distributed architecture of the AST-TDMA protocol, each vehicle needs to do its own
calculations of its next trajectory. Therefore, each vehicle will need to receive and exchange its
local data. This is done in a similar way to ATP-MAC, where omni-directional antennas are used
to broadcast (one-to-many) packets, which will allow all neighbourhood vehicles within hearing
range to receive the transmitted packets and therefore each neighbour vehicle’s local
information. This allows each vehicle to do its own trajectory calculations based on the
swarming algorithm used for formation control. As the new trajectory calculation is done in the
vehicle, then this trajectory can be implemented immediately, which for timely manoeuvring will
be an advantage.
5.4.1 AST-TDMA Packet Structure
The two types of packets required for this architecture is the Data Packet from the swarm
vehicles that includes a token, or a Token Packet if the vehicle has no data to send, see Table
5.2. The Command Packet of the ATP-MAC is not required in this protocol. The data traffic
required to send from each vehicle is its local information related to current position and
planned trajectory and will depend on the swarm formation algorithm used. Also, depending on
the application, the payload sensor data collected by a vehicle may need to be used to
determine the new trajectory in that vehicle. That is, for example, when trying to find the source
of an algae bloom the vehicle will incorporate the sensor data concentration variations into its
own trajectory calculations which will impact on the direction that that vehicle takes and
therefore indirectly influences the direction of the swarm as has been discussed in Chapter 2.6.
Payload data may also need to be sent as data separately for redundancy purposes. As a
minimum each of the neighbour vehicles localisation information is required in a time critical
manner to avoid a vehicle collision with one of its neighbours, especially at low ranges as
shown in Section 5.9.
95
The same frame structure is used as in the ATP-MAC protocol and based on the IEEE802.5
Token Ring protocol. The size and structure of these packets, are shown in Table 5.2 and are
designed as the same size as the ATP-MAC protocol which include the same Data Packet size
of 53 bytes (424 bits) and the Token Packet is 13 bytes (104 bits). Changes in Data Packet size
to incorporate payload data is investigated here and in future work the prioritising of payload
data as an alternative approach. Similar to the ATP-MAC protocol, the token in AST-TDMA
functions as a simple control packet that requires the source and destination address
information to signify the vehicle that did not have navigational information to send and the
destination vehicle which it is passing on the permission to access the channel. This is different
to the token in the IEEE802.5 where the AC (Access Control) contains the token bit that a node
can grab if it is free.
5.4.2 AST-TDMA Cycle Description
A typical single cycle is illustrated in Figure 5.3 with the change for the ATP-MAC of the
Command Packet being replaced by a Data or Token Packet, so can be defined in the same
way as a set of back-to-back transmissions and the propagation times between consecutive
vehicles. Like TDMA, each node is given a slot to transmit in. The transmission sequence
through the swarm vehicles will be initially based on LIDCA (Lowest Identifier Clustering
Algorithm [113]) so that vehicle V1 will begin the process and pass its token to the swarm
vehicle with the next lowest ID in the swarm which in Figure 5.3 is V2 and then sequentially in
ascending order of vehicle ID value. The major difference to the TDMA approach is that instead
of a fixed slot size where vehicles transmit their packets in their slot based on a clock timing, the
AST-TDMA like ATP-MAC has adaptive slot lengths where a vehicle sends its packet based on
a trigger which is the arrival of a token directed to it.
Thus, when vehicle V2 completes the reception of V1’s packet, it schedules and begins the
transmission of its Data Packet. This can occur while V1’s packet is still propagating and being
received by other vehicles in the neighbourhood. This is the spatial-temporal advantage that
these two protocols have incorporated and will be discussed in Section 5.5. When vehicle V1
receives the token from the highest ID vehicle V4 in Figure 5.3, then V1 begins the next cycle. If
there is no data ready to be transmitted in any of the vehicles then they immediately send on
just the token for the cycle to continue as a ring maintenance strategy.
Thus, similar to the ATP-MAC protocol, swarm vehicles can transmit either a Token Packet with
transmission time Ttoken or a combined Data and Token Packet with transmission time (Tdata),
depending if there is data to send. These packets are broadcast so that all vehicles can
overhear each other. It can be seen in Figure 5.3 that each vehicle’s packet is received by all
vehicles in the swarm, however it is only the vehicle with the next highest ID that gets
permission to access the medium and the right to start transmitting immediately. The rest of the
cycle time is made up of propagation times between sequenced vehicles, e.g. t12 which is the
propagation time between V1 and V2.
96
5.4.3 Cycle Time (Tcycle) Analysis
The average protocol cycle time 𝑇𝑐𝑦𝑐𝑙𝑒������� can be determined by adding the time intervals illustrated
in Figure 5.4 for a swarm with V vehicles and by generalising the equation by averaging the
propagation times between sequenced vehicles as 𝑡𝑆𝑆 ���� . The average cycle time for the AST-
TDMA protocol, 𝑇𝑐𝑦𝑐𝑙𝑒�������, includes the fixed times of sending each vehicle’s token packets or token
portion of the Data Packets (VTtoken) and the processing time that each packet takes to be
received and transmitted in a vehicle. The varying parts of 𝑇𝑐𝑦𝑐𝑙𝑒������� include the time to send the
remaining portion of the Data Packet which is included based on the number of data packets
waiting in the queue to send in a cycle (𝑁𝑑����(Tdata - Tdata)), where 𝑁𝑑���� is the expected number of
packets per cycle. The propagation times between vehicles that are in sequence, V 𝑡𝑆𝑆 ����, is also
variable based on the changing ranges between sequenced vehicles.
The time for one cycle 𝑇𝑐𝑦𝑐𝑙𝑒������� is;
𝑇𝑐𝑦𝑐𝑙𝑒������� = 𝑉𝑇𝑡𝑜𝑘𝑒𝑛 + 2𝑉𝑡𝑝𝑟𝑜𝑐𝑒𝑠𝑠 + 𝑁𝑑����𝑇𝑑−𝑡 + 𝑉 𝑡𝑆𝑆 ���� (5.2)
5.5 Using Spatial-Temporal Diversity Thus, both the protocols proposed have been designed to utilise the unique spatial-temporal
diversity of the underwater channel, which allows the protocols to use concurrent transmissions
without collisions. Figure 5.4, repeated here from Chapter 4 Figure 4.1, illustrates some of the
spatial-temporal circumstances that can occur underwater when using a four-vehicle system, V1
to V4, and has been explained briefly in Section 4.4.5. The following description of how spatial-
temporal diversity is defined in this work and operates is based on this topology and the AST-
TDMA protocol, however it also applies to the ATP-MAC protocol. Here spatial temporal
diversity is introduced due to the long propagation delay. Figure 5.4 is used as an illustrative
topology in this chapter rather than the scenarios listed that were describe in Section 4.4.5. That
is, the second scenario presented where V1 and V4, transmit at the same time will not occur
when using the adaptive protocols proposed.
The AST-TDMA and ATP-MAC protocols can permit ‘non-exclusive channel access’ and allow
concurrent transmissions in the channel, by taking advantage of the spatial-temporal diversity
experienced underwater. Figure 5.5 is a repeat of Figure 5.3, with the addition of highlighting
the non-exclusive channel access where vehicles are able to receive previous transmissions
from other vehicles while the vehicle with the allocated transmission slot is transmitting. Only
one transmitter can transmit at any time. Transmitters are allocated a slot in a cycle, and even
though these slots will vary in length and as a consequence the cycle times will vary, they are
designed so that packet collisions are avoided. The protocol therefore effectively changes the
slot size and timing of an information exchange cycle, which when vehicles are moving will
mean that the slot size will vary cycle to cycle as well.
97
Figure 5.4: Spatial-Temporal Diversity Explained. A Simple Four-Vehicle Topology
Figure 5.5: AST-TDMA: One cycle of Slot Times Based On Configuration of Figure 5.4
98
Figure 5.6: Determining Validity Of Non-Exclusive Access
The most critical timing in this protocol and the essence to its success is to ensure the
avoidance of packet collisions at a third vehicle such as V4 in Figure 5.6. That is, V2’s packet
must have finished its reception at vehicle V4 by the time vehicle V3’s packet arrives. This can
be confirmed if the transmission time of the packet that goes directly to V4 is less than the
transmission time via V3. Using the triangle inequality, the range r24 ≤ r23 + r34 needs to be
satisfied if this protocol is to work, refer to Figure 5.6.
Let the propagation delay between vehicle V2 and V3 be t23 and the transmission time of the
packet sent from V2 is T2. Then let V2 be broadcasting a package to V3 and V4, with V3 next in
sequence to broadcast its packet. To avoid a collision at the receiver of V4, the time for V2’s
packet to end reception at V3 and for V3’s packet to start its arrival at V4, which is equal to t23 +
T2 + t34, needs to be greater than or equal to the time for V2’s packet to end reception at V4,
which is t24 + T2. Thus, t24 + T2 ≤ t23 + T2 + t34 or,
t24 ≤ t23 + t34 (5.3)
The propagation delay between vehicles i & j equals the range between them divided by the
speed of sound underwater (1500m/s), i.e. tij = rij/1500. Substituting, this in (5.3) gives r24 ≤ r23 +
r34 which by the triangle inequality is true. If r24 = r23 + r34 then the vehicles are in a straight line
and the reception of the packet from V2 via r24 is completed at the time instance that the packet
from V3 via r34 arrives. Thus, even without considering processing time, which adds a small time
buffer, this would still not cause collision.
Therefore, exclusive channel access based on the transmission time of data becomes an
ineffective way to avoid collisions, unless large guard times are incorporated to take into
99
account propagation delays between all possible vehicles in an underwater network. Therefore,
non-exclusive access can occur due to the space diversity, which allows more than one
transmission-reception activity in the channel at the same time.
5.6 Conventional TDMA Protocol A conventional TDMA protocol will be used as a comparison in the evaluation of the ATP-MAC
and AST-TDMA protocols as it is a closely related protocol and is being used in current
applications of groups of AUVs [67].
The TDMA protocol is one of the fundamental time-based protocols and effectively shares the
medium by dividing the time equally among the vehicles in the network, with each vehicle
assigned its own time slot. Therefore, the time-slot size is an important parameter to determine
and is generally fixed and defined by the packet size and maximum propagation delay expected
in the network. Thus the 𝑇𝑐𝑦𝑐𝑙𝑒������� is fixed for the TDMA protocol as the slot size is fixed, Tslot. The
Tslot includes the maximum packet size (Tdata), the maximum propagation delay and a guard
time. The maximum propagation delay is dependent on topology but will be determined here as
twice the average propagation delay (𝑡𝑆𝑆���� ) between sequenced vehicles, 2 ∗ 𝑡𝑆𝑆����. This value,
2 ∗ 𝑡𝑆𝑆���� is used as it provides allowance for propagation equivalent to two-hops or propagation to
its neighbour’s neighbour. The Tslot also includes a guard time of the maximum propagation
delay, 2 ∗ 𝑡𝑆𝑆����. Therefore, the fixed slot time is;
𝑇𝑠𝑙𝑜𝑡 = 𝑇𝑑𝑎𝑡𝑎 + 4 𝑡𝑆𝑆���� (5.4)
The vehicle sequence cycle time for the TDMA protocol is simply determined by the total
number of fixed slot times and the number of vehicles in the swarm;
𝑇𝑐𝑦𝑐𝑙𝑒 = 𝑉𝑇𝑠𝑙𝑜𝑡 (5.5)
5.7 Performance Criteria As with terrestrial sensor networks, the performance criteria typically used for evaluating MAC
layer protocols include delay, throughput, fairness, and energy efficiency [73]. For an
underwater swarm sensor network (USSN) the performance criteria will focus on delay,
throughput and fairness. Energy consumption will not be considered further here as discussed
in Chapter 3, Section 3.5.6.
Swarm synchronisation has been defined as the ultimate purpose of the explicit exchange
control and/or navigational data between vehicles. Swarm synchronisation is considered here to
be the ability to maintain a successful swarm-like formation of the vehicles and means that
vehicles can successfully manoeuvre within close proximity to each other without vehicle
collisions and be able to know the overall direction of the swarm and the current mission status.
100
5.7.1 Network Delay
The first major performance criterion of the protocol is the timely exchange of information
between all vehicles in the swarm, which is a packet exchange delay across the whole network.
This is vital to the operation of a swarm as it is the trajectory update times in each vehicle that
will ensure safe and successful swarm operations so as to maintain swarm synchronisation and
the required QoS. This is referred to as the Network Delay.
A new metric, Neighbourhood Communication Cycle Period (NCCP), has been developed to
analysis this network delay that describes the time for all vehicles to be updated from all their
neighbours. This is critical from an application layer perspective as it is the full exchange of
localisation data that is needed to determine the position of closely operating vehicles and as
input into the vehicle’s own next trajectory for the AST-TDMA protocol. The maximum NCCP
allowable for the two topologies investigated in this work will be presented in Section 5.7.4.
Both protocols work on a Cycle Time, Tcycle, as illustrated in Figures 5.3. Tcycle is the MAC layer
measure for NCCP and Tcycle is equal to NCCP when all vehicles have a packet ready to send in
their allocated slot in each cycle. If this is not the case then more than one cycle will be required
to complete data exchange, which means that NCCP will be a multiple of Tcycle. Thus the first
boundary limitation for the protocol is that NCCP = Tcycle so that no vehicle misses a slot to send
data, which decreases the time efficiency of the protocol. This is evaluated in Section 5.9.
Swarm synchronisation can be maintained if each vehicle can determine and implement its new
next trajectory within the QoS boundaries. That is, the time that it takes for the application layer
to receive all its neighbours’ localisation data so as to avoid vehicle collisions and to do the next
trajectory calculation.
5.7.2 Channel Resource Utilisation and Throughput
The second major performance criterion is to evaluate the Channel Capacity Utilisation. This is
a different measure to the more traditional definition used in terrestrial applications where
minimising the Channel Utilisation is seen as a positive so that it allows the available resources
to be used for other purposes. In a single channel underwater environment, where non-
exclusive access is permitted, the useable capacity of the channel is best to be maximised so
that the limited resources can be fully utilised as the unused portion of the channel are wasted
as they cannot be used for anything else. A closely related metric is throughput which with
Channel Capacity Utilisation will be developed and explored in Chapter 6.
5.7.3 Swarm Synchronisation
In a USSN, the ability of the swarm to maintain swarm synchronisation is a critical condition that
needs to be satisfied and is a major aim of the communication. Therefore, maintaining swarm
synchronisation is a Quality of Service (QoS) parameter that will be defined here.
Swarm synchronisation is the ability of the swarm to maintain a swarm-like configuration or
swarm cohesion or more specifically that the swarm vehicles can successfully operate in close
101
proximity to each other without vehicle collisions and with control of the overall direction of the
swarm. To do this the next operating state of the network needs to be determined on a
consistent and reliable basis.
For the Bus Topology and the ATP-MAC protocol the data exchange is time-sensitive as each
vehicle has a pre-planned trajectory for the duration of the mission and the exchange is to
ensure vehicle collisions do not occur that is of concern in these closely operating groups. Data
exchange is also useful to allow the swarm to operate in a dynamic surveying mode where
variations to the pre-set trajectory can be organised across the swarm for more efficient mission
outcomes. This could mean the exchange of both localisation and/or payload data.
A swarm formation algorithm is required in the Cluster Topology to provide the control of
vehicles position within the swarm and these algorithms require exchange of localisation data
from all its neighbourhood vehicles, which is thus a priority for swarm data communication. The
AST-TDMA protocol therefore has time-critical data exchange requirements. Payload data is
required as input to the swarm formation algorithm in each vehicle and therefore may or may
not be needed for explicit exchange.
Payload data exchange is considered a secondary data communication priority for both
protocols and may be distributed depending on mission requirements and redundancy of data
due to risk of loss of a vehicle. Evaluating packet size to incorporate payload data will be
evaluated in Section 5.9.5.1 and further in Chapter 6.
The remainder of this Section will establish the QoS boundaries required for swarm
synchronisation based on the two topology options.
Figure 5.7: Potential Disturbance in Bus Topology
102
Table 5.3: NCCPhard for Vehicle Collisions Based on Disturbances in Bus
Range between vehicles: 30 m
Speed: 1 m/s Speed: 2 m/s Speed: 3 m/s
θ S⌃ NCCPhard θ S⌃ NCCPhard θ S⌃ NCCPhard
45° 1.4 m/s 27 s 45° 2.8 m/s 12 s 45° 2.8 m/s 7 s
25° 2.3 m/s 11 s 25° 4.7 m/s 4 s 25° 4.7 m/s 1.7 s
Range between vehicles: 50 m
Speed: 1 m/s Speed: 2 m/s Speed: 3 m/s
θ S⌃ NCCPhard θ S⌃ NCCPhard θ S⌃ NCCPhard
45° 1.4 m/s 47 s 45° 2.8 m/s 22 s 45° 2.8 m/s 13.7 s
25° 2.3 m/s 20.3 s 25° 4.7 m/s 8.65 s 25° 4.7 m/s 4.7 s
5.7.4 Performance Boundaries
To establish the QoS NCCP maximum time allowed to exchange data, every vehicle to every
vehicle, there are two operational boundaries, NCCPsoft and NCCPhard, which can be defined for
the metric NCCP (Neighbourhood Communication Cycle Period). The shortest time of these will
define the NCCPlimit for a specified topology arrangement.
5.7.4.1 NCCPsoft Bounds – Due To Failure
The soft bound, NCCPsoft, represents an operational ideal time interval for full exchange of each
vehicle’s data. This bound is based on vehicles ability to manoeuvre so as not to cause a
vehicle collision should a transmission or even vehicle fail. NCCPsoft is therefore determined by
a vehicle’s manoeuvrability control (M meters), and how long it takes to manoeuvre to maintain
the desired trajectory once it has received and can implement a new trajectory.
Manoeuvrability will depend on the speed that the swarm vehicles are travelling at. Based on
knowledge of the SeaVision™ AUV [86], the vehicle needs at least M = 3 m if vehicles are
travelling at 1 m/s or M = 6 m at 2 m/s to be able to manoeuvre so as to avoid a vehicle
collision. Therefore,
NCCPsoft = 3 s (5.6)
is included in Tables 5.3 to 5.6.
103
5.7.4.2 NCCPhard Bounds – Due To Vehicle Collision
The hard bound, NCCPhard, is the worst case scenario that must not be breached and is defined
as the time limit available before any vehicle collisions occur. Vehicle collisions will occur if a
vehicle is somehow disturb off its planned trajectory which may occur based on the functionality
of the vehicle (incorrect trajectory calculation, loss or failure of a vehicle) and due to
environmental factors such as water currents and obstacles. Some water currents may affect all
vehicles equally, which will not cause problems for vehicle collisions in either topology (but will
need to be taken into account for absolute positioning). Single vehicle disturbances, however,
may occur due to isolated water movements, and bouncing off obstacles such as fish or terrain
where vehicles may change direction.
The NCCPhard bound is dependent on topology, vehicle range and speed and therefore will be
evaluated separately for the Bus and Cluster topologies. These calculations will assume the
swarm is travelling at a speed S m/s and there will be a force applied to one vehicle differently
to the other vehicles in the swarm. The angle of disturbance on a vehicle by this force is
represented by θ, where a large disturbance, based on collision with an obstacle or large sea
state is represented by θ = 25° and a small disturbance, related to isolated water movements
and calmer sea states by θ = 45°.
5.7.4.3 NCCPhard Bound for Bus Topology
For the Bus Topology, the loss or failure of a vehicle within the swarm will not be expected to
cause a vehicle collision as that vehicle will simply be left behind in the structured movement of
the swarm (it will cause an additional delay for the communication around the swarm which will
be discussed in relation to the packet delay).
Should a disturbance happen to a single vehicle, in a Bus Topology, as shown in Figure 5.7,
then there are potential scenarios where vehicle collisions may occur. The time to a vehicle
colliding is an estimate of the maximum NCCPhard. This time is illustrated using a disturbance
that has occurred to one vehicle. For this example, the force on the disturbed vehicle would also
need to include an increase in speed for a collision to occur with the undisturbed vehicle, which
may not be out of the question with a force pushing on it. This speed change, S⌃ is shown in
Table 5.3. In the alternative scenario where a V shape rather than a line formation is shown, a
change of speed on the disturbed vehicle may not be necessary to still cause a collision. The
time to collision tcoll can be determined using θ, S and the variables r as the range between
vehicles and d as the distance that the vehicle would travel on its planned path. Thus:
𝑡𝑐𝑜𝑙𝑙 = 𝑟 tan𝜃𝑆
(5.7)
as 𝑑 = 𝑆 × 𝑡 and tan𝜃 = 𝑑𝑟 .
104
To avoid a collision, each vehicle will need enough time to obtain information from its
neighbours and re-calculate its trajectory with enough range to manoeuvrer out of a collision
path. Thus the NCCPhard bound, is the time to collision minus the manoeuvrability time obtained
in Section 5.7.4.1:
𝑁𝐶𝐶𝑃ℎ𝑎𝑟𝑑 = 𝑟 tan𝜃𝑆
− 𝑀 (5.8)
Table 5.4: Summary of NCCPhard and NCCPsoft Bound for Bus Topology
Vehicle Speed: 1 m/s (2 knots)
Vehicle Speed: 2 m/s (4 knots)
Vehicle Speed: 3 m/s (6 knots)
Ave Rge (m)
NCCP hard (s)
25° (large)
NCCP hard (s)
45° (small)
NCCP soft (s)
NCCP hard (s)
25° (large)
NCCP hard (s)
45° (small)
NCCP soft (s)
NCCP hard (s)
25° (large)
NCCP hard (s)
45° (small)
NCCP soft (s)
30 11 27 3 4 12 3 1.7 7 3
50 20.3 47 3 8.6 22 3 4.7 13.7 3
* Underlined values are the NCCPlimit for each range and swarm speed
Figure 5.8: Potential Disturbance in Cluster Topology
105
Table 5.3 provides some examples of NCCPhard for the Bus Topology at various swarm speeds
and ranges. Table 5.4 presents a summary of the NCCPhard and NCCPsoft bounds and
establishes the NCCPlimit’s for the Bus Topology. In general, the NCCPsoft bound is the QoS limit
however as the vehicle speed increases and vehicles are operating at shorter ranges the
NCCPhard bound will determine the NCCPlimit.
5.7.4.4 NCCPhard Bound for Cluster Topology
For the Cluster Topology as illustrated in Figure 5.2, the failure of a vehicle has a possibility of
causing a collision as neighbouring vehicles may be on a similar trajectory behind a failed unit.
Considering Figure 5.8, where a single vehicle experiences a disturbance, then the time to a
collision is the distance represented by h = 𝑟2
tan𝜃 . Thus, the time to collision tcoll for the Cluster
Topology varies from Equation 5.7 by 𝑟2 instead of r. It follows that NCCPhard bound, which is the
time to avoid a collision, can be calculated by taking away the manoeuvrability time; thus
NCCPhard bound for a Cluster Topology with a single vehicle disturbance is:
NCCPhard =𝑟2tan𝜃
𝑆 - M (5.9)
Table 5.5 provides some examples of the NCCPhard and NCCPsoft bounds for various
transmission ranges and at different speeds of operation for the Cluster Topology and similarly
to Table 5.3 establishes the NCCPlimit’s for the Cluster Topology. First item to note is that at
higher speeds and shorter ranges there are situations in which a large disturbance to one
vehicle will be impossible to recover from. That is, even if the vehicles had instantaneous
communication about the fact, it could not manoeuvre out of a collision path as indicated by the
communication limit of <M less than manoeuvrability range. When vehicles are travelling at
these higher speeds, even at the higher range of 50 m between vehicles there is not much time
for vehicles to manoeuvre out of a collision path and the NCCPhard bound is lower than the
NCCPsoft bound.
5.7.4.5 NCCP Limits
For the remainder of this work an assumption of the swarm travelling at 2 m/s and experiencing
either a large or small disturbance will be used and therefore a summary of the NCCPlimits are
presented in Table 5.6 for the two topologies and therefore for the two proposed protocols. The
TDMA protocol will be used in both topologies as a comparison.
106
Table 5.5: Hard and Soft Time Boundaries of NCCP for Cluster Topology
Vehicle Speed: 1 m/s (2 knots)
Vehicle Speed: 2 m/s (4 knots)
Vehicle Speed: 3 m/s (6 knots)
Ave Rge (m)
NCCP hard (s)
25° (large)
NCCP hard (s)
45° (small)
NCCP soft (s)
NCCP hard (s)
25° (large)
NCCP hard (s)
45° (small)
NCCP soft (s)
NCCP hard (s)
25° (large)
NCCP hard (s)
45° (small)
NCCP soft (s)
30 4 12 3 0.5 4.5 3 <M 2 3
50 8.6 22 3 2.8 9.5 3 0.9 5.3 3
* <M means that communication time is greater than manoeuvrability time of vehicles
* Underlined values are the NCCPlimit for each range and swarm speed
Table 5.6: Summary of NCCPlimit (s)
Vehicle Speed: 2 m/s (4 knots) θ = 25° - Large Disturbance θ = 45° - Small Disturbance Bus Topology Cluster Topology Bus Topology Cluster Topology NCCPlimit (s) NCCPlimit (s)
20 m 1.7 - 3 2
30 m 3 0.5 3 3
40 m 3 1.7 3 3
50 m 3 2.8 3 3
5.8 Queuing Model Analysis For both topologies, there is a time-sensitive or time-critical Data Packet that each vehicle
needs to produce and exchange with all its neighbours on a continuous and regular basis. The
packet includes information on a vehicle’s present and planned position, orientation and
velocity, which is obtained from the navigational (referred to as the vehicle’s localisation data)
and payload sensor data. Payload data may be distributed to other vehicles directly as data
within the packet or indirectly when incorporated within the swarm formation algorithm that
determines the vehicles new trajectory. In this analysis a fixed data packet size is used, and
various fixed packet sizes will be investigated in this work.
The analytical model used to abstractly evaluate the network and packet delay performance of
the two adaptive protocols is based on a queuing network model. We have assumed that
packets are generated at each node according to a deterministic process with rate λ
packets/second and the service times having a general distribution that is dependent on the
vehicle sequence cycle time, Tcycle.
The analysis will first establish the average number of data packets expected in a cycle (Nd) so
that an average cycle time (𝑇cycle�������) can be determined based on Equations 5.1, 5.2 and 5.5. The
average expected Neighbourhood Communication Cycle Period (NCCP), which is the time
required for a full exchange of data from every vehicle to every vehicle, can be determined from
107
the 𝑇cycle������� and the Packet Arrival Rate (λ pkts/s) in the MAC layer queue. The packet arrival rate
is critical in determining if there is going to be a packet in the vehicle’s queue when it is its turn
to send a packet in each cycle. If the packet arrival rate is too slow and Tcycle is based on a cycle
where not all vehicles have send a data packets, but have had to send a token, then multiple
cycles will be required to enable a full exchange. This means NCCP will increase and
introduces time inefficiencies into the exchange, which degrades performance, as the NCCP
value is the applications QoS requirement. Thus, a full cycle of packets from all vehicles in the
swarm will be referred to as a saturated cycle and the swarm size and range limits will be
evaluated first such that the criteria NCCP = Tcycle is met.
As Packet Arrival Rate in a vehicle MAC layer queue increases, then the packets in the queue
will increase. As navigational data exchange is time-limited, there is no benefit in exchanging
old data when newer data is available for exchange, as the newer data will more accurately
reflect the current location of the vehicle. Therefore a LIFO (Last In First Out) queue or packet
lifetime mechanism needs to be implemented for this type of data, which means that not all
information collected will be exchanged and old data discarded for swarm synchronisation
purposes. Thus the queue will be sorted based on the age of the packet and older packets
dropped. There maybe future application requirements that need this ‘older’ information and if
this is the case, which will not be implemented here, then each vehicle will need some
additional process to retain the information prior to discarding. This provides a secondary QoS
limitation, which will be explored in the next Section.
The limitations on system parameters such as the swarm size, the range between vehicles and
packet size will be determined to ensure the ability of the swarm to maintain swarm
synchronisation. Finally, the protocols will be evaluation against a conventional TDMA protocol.
The following are the notations and assumptions that are used in the following analysis:
• A swarm is defined as a collection of V vehicles randomly dispersed but maintaining
similar ranges between their closest neighbours. For the ATP-MAC protocol the network
has a coordinator vehicle and (V-1) swarm vehicles. For the AST-TDMA and TDMA
protocols there are V swarm vehicles.
• Each vehicle will obtain on a regular basis a new trajectory plan. This will be from the
coordinator vehicle in the ATP-MAC protocol and from its own calculations in the AST-
TDMA and TDMA protocols. The next trajectory calculations are based on a swarm
algorithm that requires exchange of data from all the other vehicles in its
neighbourhood.
• It is a single-server multi-queuing system where the sensor data packet arrival rate into
each vehicle’s queue is λ packets/second, and is independent of the other vehicles. The
total arrival rate is thus V*λ.
108
• All vehicles are assumed to have a very large buffer size compared to the arrival rate so
that packets are not lost due to buffer overflow. At this stage the queue is designed as a
Last in First out (LIFO) arrangement where only the last packet from the queue will be
transmitted and the older packets discarded, as only the most up-to-date information is
required.
• The service time of packets is based on the vehicle sequence cycle time, Tcycle, which is
dependent on the number of swarm vehicles, range between vehicles and packet size.
• Packet transmission rates are fixed at 9600 bps
• Each packet type has a fixed length in each scenario and therefore a transmission time
of Tcomm (Command), Ttoken (Token), or Tdata (Data) - refer to Table 5.2. A general packet
transmission time will be referred to as (Tx).
5.8.1 Model Parameters
The following results have been simulated using MathWorks Matlab numerical computing
environment. The base parameters used in these calculations are as detailed in Table 5.7, with
variations to range (r), packet size (L) and disturbance level (θ) tested in the following sections.
5.9 Impact of Network Delay on Swarm Size
5.9.1 Expected Data Packets per Cycle
In an average cycle there is expected to be 𝑁𝑑���� data packets available during the cycle across all
of the swarm vehicle queues, that is, Nd represents the expected number of data packets that
can be accommodated in one cycle. From Little’s Theorem, the mean number of packets N(t)
across all node queues N in a time t, can be represented as E[N(t)] = Nλt.
For the AST-TDMA and TDMA protocols, there are V swarm vehicle queues that each have an
independent packet arrival rate of λ pkts/s and therefore a total network arrival rate of Vλ pkts/s.
The time over which the number of packets is evaluated is a cycle time Tcycle and thus:
𝑁𝑑���� = E[N(Tcycle)] = Vλ𝑇𝑐𝑦𝑐𝑙𝑒������� (5.10)
Tcycle varies based on Equation 5.2 for the AST-TDMA protocol or Equation 5.5 for the TDMA
protocol and is evaluated here based on an average range between sequenced vehicles. Nd will
not be greater than the number of vehicles participating in the cycle as each vehicle sends only
one packet per cycle. Once Nd reaches V the cycle will be full and this will now be referred to as
a saturated cycle. Therefore, at saturation:
𝑁𝑑���� = V (5.11)
For the ATP-MAC protocol, there is a coordinator vehicle, which collects sensor data in a similar
way to the swarm vehicles however it does not need to exchange its data packet as it only
exchanges its Command Packet once a cycle. The Command Packet provides control of the
swarm and the protocol itself and is required to signal the start of a cycle. Thus the Command
109
Packet is not counted as part of the swarm vehicle data exchange but is required each cycle
and therefore Nd = 1 as a minimum. Otherwise, the ATP-MAC protocol has V-1 swarm vehicle
queues that each has an independent packet arrival rate of sensor data of λ pkts/s, and
therefore a total network queue rate of (V-1)λ. The average expected number of Data Packets
𝑁𝑑���� ready for exchange in one vehicle sequence cycle time Tcycle (from Equation 5.1), is:
𝑁𝑑���� = E[N(Tcycle)] = (V-1)λ𝑇𝑐𝑦𝑐𝑙𝑒������� + 1 (5.12)
For a fully saturated cycle, which means that each swarm vehicle has sent a Data Packet in that
cycle, 𝑁𝑑���� = (V-1) + 1 = V, equivalent to Equation 5.11. As with the AST-TDMA protocol, Nd will
never be greater than the number of vehicles participating in a cycle.
Sensor data collected at the application layer in each of the vehicles is provided to the MAC
layer queue ready for exchange with other vehicles in the swarm. The arrival rate of this data in
each vehicle is λ pkts/s, which is an important parameter for these time-based protocols as it will
determine if there is a packet ready to send when the protocol gives permission to that vehicle
to access the medium and send its data. Figure 5.9 shows how the expected number of data
packets that can be accommodated in an exchange cycle changes for the three protocols with
increasing arrival rates for a 15-vehicle and a 5-vehicle swarm based on Equation 5.10 and
5.12.
Taking the 5-Vehicle AST-TDMA protocol (bottom blue line) as an example, at λ = 0.36 pkts/s
there is expected to be only one packet available in each cycle from the 5-vehicle queues.
Below this arrival rate, there may be some cycles that do not include any data packets from all
of the vehicles and therefore in these cycles only token packets are sent from all vehicles to
ensure the cycle is maintained. Then at λ = 0.96 pkts/s, Nd = 2 where 2 vehicles are expected to
have a packet to send in each cycle. As the arrival rate of packets into the MAC queue
increases, the inter-arrival time (1𝜆) decreases until it equals the cycle time, which means that a
packet will arrive in each vehicle queue in time for each cycle. This can be seen if Equation 5.10
is re-arranged to evaluate 𝑇𝑐𝑦𝑐𝑙𝑒������� and Equation 5.11, Nd = V, is substituted, such that at
saturation:
𝑇𝑐𝑦𝑐𝑙𝑒������� = 𝑁𝑑���� 𝑉𝜆
= 𝑉𝑉𝜆
= 1𝜆 s (5.13)
Table 5.7: Base Parameters used in Initial Analysis
System Parameters Packet Parameters (refer Sections 5.3 & 5.4) r 50 m Lcomm Command: 121 bytes (968 bits)
V 2 to 40 vehicles Ltoken Token: 13 bytes (104 bits)
R 9600 bps Ldata Data: 83 bytes (664 bits)
λ Deterministic (pkts / s)
θ 45° (Small disturbance)
110
At an Arrival Rate of 1 packet per vehicle per cycle or above, the number of packets that can be
accommodated in the swarm reaches its maximum, which equals the total number of vehicles in
the swarm. For the 5-Vehicle Swarm using AST-TDMA this point is λ = 2.33 pkts/s, where the
cycle became a full cycle or a saturated cycle. It can be seen in Figure 5.9 that saturation
occurs for all three protocols; AST_TDMA, ATP-MAC and TDMA protocols at 15 packets and 5
packets respectively or when 𝑁𝑑���� = V (Equation 5.11).
The TDMA protocol reaches its saturation at a much lower Packet Arrival Rate (or higher inter-
arrival time) as the time slots are much longer due to the guard times required by the TDMA
protocol and therefore cycle times will be longer as shown in Figure 5.11. The ATP-MAC
protocol saturation occurs at a slightly lower Packet Arrival Rate than for the AST-TDMA
protocol due to the slightly longer cycle times and the requirement of the Command Packet that
impacts on the ATP-MAC Tcycle. The Command Packet is also illustrated in the expected
number of packets for an ATP-MAC cycle at very low Packet Arrival Rates where is can never
be less than 1, based on Equation 5.12 and seen in Figure 5.9 for values below λ = 0.25 pkts/s
for the 5-Vehicle swarm.
Figure 5.9: Average Expected Number of Packets Serviced per Cycle for Increasing Packet Arrival Rate at 50 m. Comparison of the TDMA, ATP-MAC and AST-TDMA
Protocols and 5 or 15 Vehicle Swarm
0.5 1 1.5 2 2.50
5
10
15
Packet Arrival Rate per Vehicle (pkts/s)
Exp
ecte
d N
umbe
r of P
acke
ts
Ser
vice
d pe
r cyc
le (p
kts)
15-Vehicles TDMA15-Vehicles ATP-MAC15-Vehicles AST-TDMA5-Vehicles TDMA5-Vehicles ATP-MAC5-Vehicles AST-TDMA
Arrival Rate when a cycle has a full set of Data Packets
111
Figure 5.10: Packets available in each vehicle per cycle at various Packet Arrival Rates in a 5-Vehicle Swarm at 50 m
Figure 5.10 illustrates the expected number of data packets predicted to arrive in a swarm
vehicle per cycle for the different protocols used using a 5-Vehicle Swarm. For each of the
protocols the Packet Arrival Rate of 1 packet per cycle per vehicle (pkts/Tcycle/V) (black
horizontal line) corresponds to the Packet Arrival Rate of a saturated cycle (of 5 packets) in
Figure 5.9. This is the minimum Packet Arrival Rate required for a particular size of swarm to
obtain QoS level. Below the minimum Packet Arrival Rate, some vehicles in the swarm may
have a packet ready to send but not all vehicles will, and this can be seen by the staircase
arrangement in Figure 5.9 where there are an increasing number of packets that can be
accommodated in a cycle.
As Packet Arrival Rate increases above this minimum, Figure 5.10 illustrates the rate at which a
vehicle will expect to obtain a second packet in the MAC queue in a cycle and as Packet Arrival
Rate increases further the vehicle queue will begin to grow. As discussed the navigational data
exchange is time-limited and therefore a LIFO (Last In First Out) queue has been initially
implemented. Following this approach, then the maximum Packet Arrival Rate for which all
packets entering the queue are exchanged will be up to the arrival of a second packet in the
queue in a cycle. Figure 5.10 shows the Packet Arrival Rate for each of the protocols for when a
second packet is received in the cycle, which means that the older packet will be discarded.
Using the AST-TDMA example for 5-Vehicles, the expected number of packets predicted in
each vehicle per cycle increases to 2 packets at λ = 4.7 pkts/s.
As discussed in Figure 5.9, the Packet Arrival Rate for the TDMA protocol is much lower
compared with the new protocols for when one packet and then the second and subsequent
packets arrive in the queue per cycle due to the larger slot lengths and cycle time. The
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50
1
2
3
4
Packet Arrival Rate per Vehicle (pkts/s)
Pa
cke
t ava
ilab
le p
er
Cyc
le
pe
r V
eh
icle
(p
kts/
Tcy
cle
/V)
5-Vehicle TDMA5-Vehicle ATP-MAC5-Vehicles AST-TDMA
Arrival Rate when Packets will be discarded
112
difference in Packet Arrival Rate between one and two packets arriving in a cycle for the TDMA
protocol is smaller and at lower rates which means that the inter-arrival time is larger giving it
more range in the arrival rate of data which would give more flexibility to the application layer
and the collection of this data. Yet because of the large guard times the throughput is expected
to be smaller and this will be investigate further in Chapter 6.
5.9.2 Cycle Time and Network Saturation
The cycle time (Tcycle) described in Equations 5.1, 5.2 and 5.5 for the ATP-MAC, AST-TDMA
and TDMA protocols respectively, is illustrated in Figure 5.11 for a 15-Vehicle and 5-Vehicle
Swarm. The increasing expected number of data packets per cycle seen in Figure 5.9 up to
cycle saturation can also be seen in the Cycle Times for the AST-TDMA and ATP-MAC
protocols in Figure 5.11 where Tcycle increases until it reaches saturation when Tcycle becomes a
constant value. This constant Tcycle above cycle saturation occurring here as a constant average
range between sequenced vehicles is used in these calculations. In simulation, when the range
may vary between vehicles in a moving swarm, this will not be the case, which means that there
may be some sensitivity around cycle saturation that may need to be taken into account.
It can be seen in Figure 5.11 (as in Figure 5.9) that for the smaller swarm size of 5-Vehicles, the
saturation occurs at a higher packet arrival rate than in the 15-Vehicle Swarm, as the cycle time
for only 5 vehicles is much lower than the cycle time for a 15-Vehicle Swarm. It is not obvious in
Figure 5.11 when saturation occurs in the TDMA protocol as Tcycle is constant irrespective of
Packet Arrival Rate, which is due to the fixed slot sizes, as discussed in Equations 5.4 and 5.5.
Cycle time for the TDMA protocol is however significantly higher than the two new protocols in
each of the swarm sizes shown and this reflects the significantly lower packet arrival rate that
can be accommodated when using TDMA shown in Figure 5.9.
Figure 5.11: Comparison of Cycle Time, for the three protocols with a 5-Vehicle and 15-Vehicle Swarm at 50m
0 0.5 1 1.5 2 2.50
0.5
1
1.5
2
2.5
Packet Arrival Rate per Vehicle (pkts/s)
Cyc
le T
ime:
Tcy
cle
(s)
15-Vehicles TDMA15-Vehicles ATP-MAC15-Vehicles AST-TDMA5-Vehicles TDMA5-Vehicles ATP-MAC5-Vehicles AST-TDMA
Arrival Rate when Cycle becomes Saturated
113
The cycle time for the AST-TDMA protocol is the lowest of the three protocols and lower than
the ATP-MAC due to the influence of the size of the Command Packet. The ATP-MAC
Command Packet is larger and more time consuming in each cycle than an ordinary swarm
vehicle’s Data Packet which Equations 5.1 and 5.2 can explain. For a 15-Vehicle Swarm, the
cycle times are 1.29 s, 1.32 s and 2.66 s for the AST-TDMA, ATP-MAC and TDMA protocols
respectively. The lower the cycle time, the quicker the cycles, which means the opportunity to
send more data more regularly or, in reverse, a longer cycle time means that the protocol
cannot transmit as many packets over the same operational period. That is, the adaptive
protocols will have an advantage over the TDMA protocol, which will have a larger queue, for
the same arrival rates. When packet arrival rates increase further, the queue in each vehicle will
grow and information will not be able to be exchanged.
5.9.3 Neighbourhood Communication Cycle Period (NCCP)
The Neighbourhood Communication Cycle Period, NCCP, is determined by the number of
cycles required to enable the full exchange of data or when the cycle is saturated. Therefore,
the fraction (𝑉−1)(𝑁𝑑−1)
for the ATP-MAC protocol and 𝑉𝑁𝑑
for the AST-TDMA and TDMA protocols,
which is the ratio of the number of vehicles participating in the cycle to the expected number of
packets from the swarm vehicles per cycle, gives the number of cycles required before full
exchange occurs. For the AST-TDMA and TDMA protocols there are simply V swarm vehicles
and Nd expected data packets exchanged per cycle whereas for the ATP-MAC protocol there is
V-1 swarm vehicles and Nd – 1 expected packets as the Command Packet is not counted in the
expected number of packets exchanged. The result for the three protocols is that they are
equivalent:
ATP-MAC: 𝑁𝐶𝐶𝑃 = (𝑉−1)(𝑁𝑑−1)
𝑇𝑐𝑦𝑐𝑙𝑒 �������� = 𝑉𝑁𝑑
𝑇𝑐𝑦𝑐𝑙𝑒 �������� (5.14)
AST-TDMA & TDMA: 𝑁𝐶𝐶𝑃 = 𝑉𝑁𝑑
𝑇𝑐𝑦𝑐𝑙𝑒������� (5.15)
Figures 5.12, 5.13 and 5.14 demonstrate the relationship between NCCP, Tcycle and the number
of packets that would be discarded per vehicle if a LIFO queue arrangement was implemented
for the AST-TDMA, ATP-MAC and TDMA protocols respectively operating in a 5-Vehicle Swarm
at an average range of 50 m between vehicles.
When the cycle is fully saturated and Nd = V, 𝑇𝑐𝑦𝑐𝑙𝑒������� is at its maximum (Figure 5.11), and in
addition, as shown in Figures 5.12, 5.13 and 5.14, 𝑁𝐶𝐶𝑃 = 𝑇𝑐𝑦𝑐𝑙𝑒�������, and NCCP will be at its
minimum at this point. That is, NCCP becomes smaller as the numbers of cycles needed to
complete a full exchange of data packets between all swarm vehicles decreases. As limiting
NCCP is a QoS requirement for the network, then the minimum Packet Arrival Rate, λmin, is a
critical parameter to be achieved for the performance of the protocols.
114
Figure 5.12: AST-TDMA protocol: NCCP, Tcycle and Packet Discards for 5-Vehicle Swarm at 50 m
Figure 5.13: ATP-MAC protocol: NCCP, Tcycle and Packet Discards for 5-Vehicle Swarm at 50 m
Figure 5.14: TDMA protocol: NCCP, Tcycle and Packet Discards for 5-Vehicle Swarm at 50 m
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50
0.5
1
1.5
2
2.5
Packet Arrival Rate per Vehicle (pkts/s)
NCCP (s)Tcycle (s)Packets discards per vehicle (pkts)
1
2
NCCP(s)
0
Cycle Saturation
PacketDiscard
Rate(pkts)
Packet Discards Every Cycle
Tcycle(s)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50
0.5
1
1.5
2
2.5
Packet Arrival Rate per Vehicle (pkts/s)
NCCP (s)Tcycle (s)Packets discards per vehicle (pkts)
0
1
2PacketDiscard
Rate(pkts)
Tcycle(s)
NCCP(s)
Packet Discards Every Cycle
Cycle Saturation
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50
0.5
1
1.5
2
2.5
Packet Arrival Rate per Vehicle (pkts/s)
NCCP (s)Tcycle (s)Packets discards per vehicle (pkts)
0
1
2PacketDiscard
Rate(pkts)
NCCP(s)
Tcycle(s)
Packet Discards Every Cycle
Cycle Saturation
115
As a note, it was not possible to detect in Figure 5.11 when the cycle time for the TDMA
protocol became saturated, which can now be identified in Figures 5.14 at just above λ = 1
pkt/s. This correlates with the results of Figure 5.10 that confirm at just above λ = 1 pkt/s, that
there was a packet available in each vehicle in each cycle for the TDMA protocol.
5.9.4 Minimum Packet Arrival Rates
The minimum packet arrival rate, λmin, plays a significant role in the performance of the new
adaptive protocols and therefore will be defined here. The λmin will change based on the
network parameters of swarm size, range between sequenced vehicles, packet size and data
rate. As data rate has been fixed this will not be considered further. The λmin ensures that each
vehicle has a packet of data to send in each cycle and is determined at the time when
𝑇𝑐𝑦𝑐𝑙𝑒������� reaches its maximum which is when Nd = V and NCCP first equals 𝑇𝑐𝑦𝑐𝑙𝑒�������. Thus
rearranging Equation 5.13, gives:
𝜆𝑚𝑖𝑛 = 1𝑇𝑐𝑦𝑐𝑙𝑒
pkts/s (5.16)
The minimum packet arrival rate thus defines the minimum NCCP for that swarm configuration,
and NCCP will remain unchanged as packet arrival rate increases above this minimum. The
minimum Packet Arrival Rate is shown in Figures 5.15 for the AST-TDMA, ATP-MAC and
TDMA protocols for increasing sizes of swarms from 2-Vehicles to 40-Vehicles.
For low numbers of vehicles in a swarm the packet arrival rates are high meaning that the inter-
arrival time that packets need to arrive in the MAC layer queue to maintain saturation is very
small as cycle time will be very short. As the number of vehicles in a swarm increases, the
cycle time will increase and therefore the inter-arrival time between packet arrivals in the queue
can increase which is the same as the arrival rate decreasing as is demonstrated in each of
these figures.
Figure 5.15: Comparison of Minimum Packet Arrival Rate for Increasing Swarm sizes at 50 m
0 5 10 15 20 25 30 35 400
1
2
3
3.5
Number of Vehicles in Swarm (V)
Min
imum
Arri
val R
ate
(pkt
s/s)
AST-TDMAATP-MACTDMA
116
It can be seen that the TDMA protocol has a much lower packet arrival rate for each size of
swarm which indicates that it will be more limited in its ability to handle as many packets per
second compared with the two new protocols. This leads to poorer data throughput that will be
seen in Section 6.4.
As packet arrival rate increases however the MAC layer queue begins to grow. At a packet
arrival rate λ2, there will be exactly two packets arriving in each vehicle in each cycle and
therefore using a LIFO queue the older packet will be discarded each cycle. This occurs when
𝜆 ∗ 𝑇𝑐𝑦𝑐𝑙𝑒������� = 2 or rearranged 𝜆2 = 2𝑇𝑐𝑦𝑐𝑙𝑒
pkts/s which is illustrated in Figures 5.12, 5.13 and 5.14.
Figure 5.16: AST-TDMA 5-Vehicle Swarm at 50 m showing Number of Cycles per NCCP and between packet discarded
Figure 5.17: AST-TDMA 5-Vehicle Swarm at 50 m showing Number of Packets queued and discarded
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50
1
2
3
4
5
Packet Arrival Rate per Vehicle (pkts/s)
Num
ber o
f Cyc
les
Number of Cycles per NCCPNumber of Cycles between packet discards
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50
0.5
1
1.5
2
Packet Arrival Rate per Vehicle (pkts/s)
Num
ber o
f Pac
kets
(pkt
s)
Queued Packets per CyclePackets discarded per vehicle Packets discarded per vehicle (every cycle)
117
In between the λmin and λ2 a vehicle will begin to build a queue but will not have packets to
discard a packet every cycle. Instead it may take several cycles in a vehicle before a discard will
occur. This is illustrated in Figure 5.16 with the number of cycles between packet discards for a
5-Vehicle swarm using the AST-TDMA protocol. For this scenario the number of cycles for a
packet to be discarded at 2.33 pkts/s, just above cycle saturation, was 2650 cycles, which
based on a cycle time of 0.4 s (Figure 5.11) would mean a packet, lost every 1060 s (17 min 40
s) per vehicle. It can be seen by the arrival rate of 3 pkts/s that a packet will be discarded
approximately every 3.5 cycles or 1.4 s which will be more likely to impact on the swarm
synchronisation. This will be investigated further in Chapters 6 and 7.
As there are several cycles between packet discards from λmin to λ2 there will be between 0 and
1 packet discarded per cycle as illustrated in Figure 5.17. Figure 5.17 also includes the
expected number of packets to arrive in a vehicles queue per cycle. This value minus the one
packet that is sent each cycle determines the packet discards per cycle.
Due to the application areas defining different topology arrangements, the remaining analysis
will now focus on these two categories; Non-Time Critical Mission (Bus Topology) and Time
Critical Mission (Cluster Topology).
5.9.5 Determination of Maximum Swarm Size
In this work, swarm synchronisation has been used to determine QoS boundary, which is
determined by the ability of the protocol to disseminate localisation data (NCCP) to the
Coordinator Vehicle in the ATP-MAC protocol and amongst all swarm vehicles in the AST-
TDMA protocol. The priority for a swarm is to ensure that it maintains swarm synchronisation
and cohesion so that vehicles do not collide or get lost from the swarm. The NCCP limits have
been analysed in Section 5.7, which provide these application QoS boundaries. Using both the
small and large disturbance scenarios and focusing first on an average range between
sequence vehicles of 50 m, the NCCPlimit for the Cluster and Bus Topologies, refer to Table 5.6,
is either 3 s or 2.8 s as shown in Figures 5.18 and 5.19.
A comparison of the NCCP time at λmin to the NCCPlimit for the Bus and Cluster Topologies
respectively, for increasing numbers of vehicles in a swarm is presented in Figures 5.18 and
5.19 to establish the maximum number of vehicles that can be supported by each of the
protocols to maintain swarm synchronisation in the different topologies they are designed for.
The TDMA protocol is used as a comparison in each of the topologies. The maximum number
of vehicles is determined where the NCCP for the vehicle crosses the NCCPlimit.
Thus based on this analytical analysis using the operational QoS boundary determined in
Section 5.7.4 and using the small and large disturbance model values, the maximum number of
vehicles that the protocols can support are summarised in Table 5.8 for the 50 m case. In
comparison to the TDMA protocol, it can be seen that the AST-TDMA and ATP-MAC protocols,
under ideal and similar conditions, allow a more densely operating swarm of vehicles.
118
Figure 5.18: Determining limit to the Number of Swarm Vehicles using Bus Topology at 50 m
Figure 5.19: Determining limit to the Number of Swarm Vehicles using Cluster Topology at 50 m
The very large difference in number of vehicles that can be supported using TDMA protocol,
which is only approximately half the vehicles, is predominately related to the fixed slot times that
need to include a large buffer to ensure packets do not collide in highly mobile environments,
which create wasted cycle time where packet exchange cannot occur. Reducing this guard time
would increase the number of vehicles that the protocol could handle, however it may result in
higher packet errors. In any case, this preliminary result shows the major advantages that the
adaptive protocols can bring to the underwater environment.
5 10 15 20 25 30 35 400
1
2
3
4
5
Number of Vehicles in Swarm (V)
NC
CP
(s)
TDMAATP-MAC
NCCPlimit for large or small disturbances (s)
5 10 15 20 25 30 35 400
1
2
3
4
5
Number of Vehicles in Swarm (V)
NC
CP
(s)
TDMAAST-TDMA
NCCPlimit for large disturbances (s)
NCCPlimit for small disturbances (s)
119
There are only minor differences in the number of vehicles the ATP-MAC and AST-TDMA
protocols can support in their particular applications. For the Bus Topology there is no
difference in NCCPlimit for the large and small disturbance scenarios and therefore the results
are the same as illustrated in Figure 5.18. For the Cluster Topology, Figure 5.19, the large
disturbance scenario requires a lower NCCPlimit and therefore means that the protocol can
support a fewer number of vehicles with a slightly higher λmin pkts/s.
These results are limiting as to allow the maximum number of vehicles in the swarm, the packet
arrival rate needs to be fixed and equivalent to the inverse of the NCCPlimit, that is λmin must
equal 1𝑇𝑐𝑦𝑐𝑙𝑒
= 1𝑁𝐶𝐶𝑃𝑙𝑖𝑚𝑖𝑡
. Take for example the AST-TDMA protocol, Figure 5.19, and using the
small disturbance environment, the maximum number of vehicles that can be supported in a
swarm is 35-Vehicles when the Packet Arrival Rate is λmin = 0.33 pkts/s, which is illustrated
again in Figure 5.20. Should the Packet Arrival Rate change within each vehicle the results for
the 35-Vehicle swarm will change. That is, if λ grows in each vehicle while maintaining a 35-
Vehicle swarm, the Tcycle remains unchanged and therefore NCCP also, however, packets will
begin to build up in each vehicles queue and will be discarded and thus lost from circulation as
was shown in Figure 5.17. If, however, λ reduces which will mean that vehicles will not have a
packet to send in each cycle, Tcycle will reduce and therefore NCCP will grow as seen in Figure
5.12 and thus a 35-Vehicle swarm will not be able to be maintained as the NCCPlimit will have
been breached.
Alternatively, examining what occurs when the number of vehicles in the swarm is reduce to say
17-Vehicles which has its λmin = 0.66 pkts/s, see Figure 5.20, but using a similar Packet Arrival
Rate of the 35-Vehicle swarm of say λ = 0.35 pkts/s. Figure 5.20 shows that for the 17-Vehicle
swarm the Tcycle would be less than its saturated value and therefore NCCP increases. At this λ
the 17-Vehicle NCCP will still be just under the NCCPlimit for a small disturbance and therefore
would be acceptable. It would however be sending several cycles with Token Packets that
would mean unnecessary transmissions.
Table 5.8: Maximum Number of Vehicles that can be supported in Small Disturbance Model at 50 m
Bus Topology θ = 25° - Large Disturbance θ = 45° - Small Disturbance ATP-MAC TDMA ATP-MAC TDMA Maximum Swarm Size Maximum Swarm Size
50 m 34 16 34 16
Cluster Topology θ = 25° - Large Disturbance θ = 45° - Small Disturbance AST-TDMA TDMA AST-TDMA TDMA Maximum Swarm Size Maximum Swarm Size
50 m 32 15 35 16
120
Figure 5.20: AST-TDMA NCCP, NCCPlimit and various Packet Arrival Rates
Thus, the packet arrival rate in the MAC queue is a critical parameter of the network to
determine cycle saturation. Once λmin is obtained then a balance between Data Packet Size,
range between sequenced vehicles, rss, and the number of vehicles in the swarm, V, is required
based on Equations 5.2 and 5.15 for the AST-TDMA protocol such that;
𝑁𝐶𝐶𝑃𝑙𝑖𝑚𝑖𝑡 = 𝑇𝑐𝑦𝑐𝑙𝑒 = 𝑉 �𝐿𝑑𝑎𝑡𝑎9600
� + 𝑉 � 𝑟𝑠𝑠1500
� + 2𝑉𝑡𝑝𝑟𝑜𝑐𝑒𝑠𝑠 (5.17)
In the next sections the analysis will focus on the limitation to the number of vehicles that the
protocols can support in a swarm, based on changes in the network parameters or packet size
and average range between sequenced vehicles. Chapter 6 will investigate the options to
improve the performance of the AST-TDMA protocol to minimise packet loss while maintaining
good channel utilisation.
5.9.5.1 Variations due to Packet Length
Variations in the Data Packet length has a direct impact on the cycle time in each of the
protocols as can be seen in Equations 5.1, 5.2 and 5.5 due to the variation in transmission time
required. This will affect the two new adaptive protocols more directly as the slot lengths are
directly based on transmission times between sequenced vehicles whereas for the TDMA
protocol, the transmission time is incorporated in the sizing of the fixed slot length and plays a
less significant role in the calculations of the slot length than propagation delay. The ATP-MAC
protocol has another data based packet, its Command Packet, which will be held at a constant
value of 968 bits even though it would be expected that small changes in packet length may be
needed especially with increasing numbers of swarm vehicles that it is controlling.
This investigation is to evaluate the influence that the sensor data packet length has on NCCP,
as the number of vehicles in the swarm increases, for the Bus Topology is shown in Figure 5.21
and for the Cluster Topology in Figure 5.22, at 50 m average range between sequenced
0 0.33 0.66 0.99 1.320
1.5
3
5
Packet Arrival Rate per Vehicle (pkts/s)
NCCP 35-VehiclesTcycle 35-VehiclesNCCP 17-VehiclesTcycle 17-VehiclesPacket Discards per cycle 35-VehiclesPacket Discards per cycle 17-Vehicles
Tcycle(s)
NCCP(s)
0
PacketDiscard
Rate(pkts)
1
2
3
NCCPlimit for small disturbances (s)
121
vehicles. The Ldata of 424 bits has been the standard length used, which incorporates just the
navigational sensor data, as discussed in Section 5.3 and 5.4, and this is being compared to
larger packet lengths based on a combination of navigational data and payload data and
number of packets of this information included as detailed in Table 5.9.
The QoS boundary limits, defined in Section 5.7, are used to establish the maximum allowable
swarm vehicle numbers with changing packet length. In these figures, the NCCPlimit for large
disturbances and small disturbances has been used to evaluate the maximum number of
vehicles that the protocol can support in each of the scenarios. The large and small disturbance
limit for the Bus Topology (Figure 5.18) at an average range of 50 m is the same value of 3 s
and relates to the NCCPsoft bound that is based on the manoeuvrability of the vehicles time to
recover following a failure rather than the potential of vehicle collisions. This is not the case for
the Cluster Topology (Figure 5.19) where for large disturbances the NCCPlimit is reduced to 2.8 s
as vehicle collisions become possible.
As packet length increases, so does NCCP for each swarm size, as is expected due to the
impact that packet length has on the transmission time. The Figures 5.21 and 5.22 illustrate that
the adaptive protocols have a more dramatic reduction in numbers of vehicles supported than
the TDMA protocol due to the reliance on the transmission time to determine slot length and the
higher number of vehicles being supported, which is due to the increasing difference in NCCP
that occurs with the additional transmission times and the larger swarm sizes being handled. As
packet length increases, these results illustrate the effect of not only the additional transmission
time per slot but the additional transmission times that are needed with increasing numbers of
slots required with increasing number of vehicles in the swarm. A note regarding the TDMA
protocol is that the guard time is unchanged, as it relies on range between swarm vehicles and
therefore propagation time only, and that slot lengths are also only increased due to
transmission times as packet size increases.
Thus for both topologies, the new adaptive protocols out performs the TDMA protocol in terms
of the number of vehicles that can be supported to maintain swarm synchronisation.
Table 5.9: Packet Size Determination
Header & Footer (bytes/bits) 13 / 104 13 / 104 13 / 104 13 / 104 Navigational Data (bytes/bits) 40 / 320 40 / 320 2*40 / 640 2*40 / 640
Payload Data (bytes/bits) 30 / 240 2*30 / 480 Total Packet Size Used
(bits) 424 664 744 1224
122
Figure 5.21: Maximum Number of Swarm Vehicles in Bus Topology with Changes in Packet Size at 50 m
Figure 5.22: Maximum Number of Swarm Vehicles in Cluster Topology with Changes in Packet Size at 50 m
0
5
10
15
20
25
30
35
40
424 664 744 1224
Max
imum
Num
ber
of V
ehic
les
in
Swar
m (V
)
Packet Size (bits)
TDMA: Large and Small Disburbances ATP-MAC: Large and Small Disturbances
NCCPlimit used = 3
0
5
10
15
20
25
30
35
40
424 664 744 1224
Max
imum
Num
ber
of V
ehic
les
in
Swar
m (V
)
Packet Size (bits)
TDMA: Small Disturbance AST-TDMA: Small Disturbance TDMA: Large Disturbance AST-TDMA: Large Disturbance
NCCPlimit used = 3 s
123
5.9.5.2 Variations due to Range between Sequence Vehicles
Reducing the range between sequenced vehicles means lower propagation delays and
therefore lower cycle times, Tcycle, as can be seen in Equations 5.1, 5.2 and 5.5. Having lower
Tcycle means that NCCP at cycle saturation will also be lower for lower ranges. Thus for the
same number of vehicles in a swarm and using the same size packet (Data Packet of 424 bits)
at a lower range, the packet arrival rate in the queue will be higher as packets can be sent more
frequently from each vehicle.
Swarms where the average range between sequenced vehicles of 20 m, 30 m, 40 m and 50 m
are compared by using the QoS NCCP boundaries determined in Section 5.7 to establish the
maximum NCCP allowable when cycle is saturated in each scenario. Figure 5.23 illustrates the
variations in the maximum number of vehicles that can be supported by the ATP-MAC and
TDMA protocols for a Bus Topology and Figure 5.24 illustrates the variations between the AST-
TDMA and TDMA protocols for a Cluster Topology.
For the Bus Topology of Figure 5.23, there is no difference in the NCCPlimit for small or large
disturbances and for the ranges of 30 to 50 m inclusive. This means that the maximum
allowable swarm size for these ranges will not change under the different disturbance situations.
As the ranges increase however there is a gradual decline in the number of vehicles that can be
supported using both the ATP-MAC and TDMA protocols. This decline is due to the additional
propagation time added to each slot in both protocols for the extra distance. This is larger for
the TDMA protocol as the guard times are dependent on propagation times. The affect of this is
greater than may first appear because while range increases the number of vehicles in the
swarm reduces which means there are less slots that need to be accommodated which will
reduce the NCCP in its own right.
In the 20 m case the NCCPlimit for small disturbances is based on the same boundary as for the
higher ranges and therefore there is a small increase in the number of vehicles that both
protocols can support for the longer ranges. This is not the case when large disturbances need
to be accounted for in a swarm with sequenced vehicles at 20 m range. In this case, the
NCCPlimit is based on the NCCPhard bound or where vehicle collisions may occur which becomes
more likely in densely operating swarms. Here the maximum number of vehicles that the
protocols can support substantially reduces, so that there is sufficient time for information
exchange to occur to allow vehicles to move out of collision paths.
The ATP-MAC protocol is able to support larger swarm in the Bus Topology than the TDMA
protocol.
In a Cluster Topology where a swarm is operating at 20 m range between sequenced vehicles
in an environment with large disturbances on a vehicle, the communication system cannot
respond fast enough to avoid a vehicle collision based on the manoeuvrability of vehicles used
in this work (in the order of 3 m). That is, for situations where large disturbances on vehicles are
possible, the ability of the communication network to be able to inform another vehicle that a
124
vehicle collision is imminent and for that vehicle to then take evasive action is not sufficient.
Thus, clustered swarms in large disturbance environments need to operate at over 20 m
ranges. For the same range of 20 m but with the possibilities of small disturbances, the
NCCPlimit is 2 s and this allows a maximum of 30-vehicle swarm when running with AST-TDMA
and only a 20-vehicle swarm for the TDMA protocol as shown in Figure 5.24.
For higher ranges, above 30 m the small disturbance environment has an NCCPlimit of 3 s and
as with the Bus Topology the maximum number of vehicles that the protocols can support in a
swarm decreases as range increases due to the increasing propagation delays. As the TDMA
protocol is not directly affected by the propagation delay differences between vehicles as the
propagation delay values are incorporated into the guard times to determine the fixed slot
lengths which is independent of configuration. Thus, the TDMA protocol can support exactly the
same maximum number of vehicles in a swarm irrespective of if it is operating in a Cluster or
Bus Topology. The operations of the adaptive protocols in their respective application topology
configurations is very similar with the only difference of AST-TDMA showing at the lower range
of 30 m that it can support 41-vehicles verses the 40 that the ATP-MAC can support.
The situation is very different for when large disturbances are possible in the operating
environment of the swarm. This has already been noted with the very closely operating vehicles
of 20 m where there is insufficient time for the communication network to respond to imminent
collisions. The effect of this is seen by the low values of NCCPlimit that increase from a low of 0.5
s, 1.7 s to 2.8 s based on the ranges 30 m, 40 m to 50 m respectively. For the AST-TDMA
protocol the number of vehicles has to reduce to remain below the NCCPlimit, from allowing from
7-vehicles at 30 m to 32-vehicles at 50 m. This is a dramatic increase in vehicle numbers which
reflects the impact of potential vehicle collisions particularly at lower ranges but also in addition
the impact that propagation delay will have within the operations of the protocol itself. This
additional impact of propagation delay can be seen more clearly when comparing the TDMA
protocol that does not have the direct impact of propagation delay and therefore the change in
maximum number of vehicles that the TDMA protocol can support is lower, rising from 4-
vehicles at 30 m to 15-vehicles at 50 m.
Thus the determination of the maximum number of vehicles that each protocol can support to
maintain swarm synchronisation with changes in range is affected firstly by the NCCPlimit and
particularly by the speed of communication to avoid vehicle collisions at shorter ranges. The
impact that propagation time has on the protocols is more complex as it is a combination of both
the increase in time as range increases but also by the number of vehicles in the swarm as this
determines the number of slots per cycle and therefore cycle time.
125
Figure 5.23: Maximum Number of Swarm Vehicles in Bus Topology for Increasing Range
Figure 5.24: Maximum Number of Swarm Vehicles in Cluster Topology for Increasing Range
0
5
10
15
20
25
30
35
40
45
50
20 30 40 50
Max
imum
Num
ber
of V
ehic
les
in
Swar
m (V
)
Range between Sequenced Vehicles (m)
TDMA: Small Disturbance AST-TDMA: Small Disturbance TDMA: Large Disturbance ATP-MAC: Large Disturbance
NCCPlimit used (s)
3 s 3 s 3 s
0
5
10
15
20
25
30
35
40
45
50
20 30 40 50
Max
imum
Num
ber
of V
ehic
les
in
Swar
m (V
)
Range between Sequenced Vehicles (m)
TDMA: Small Disturbance AST-TDMA: Small Disturbance TDMA: Large Disturbance AST-TDMA: LargeDisturbance
NCCPlimit used (s)
126
5.10 Conclusion This Chapter has introduced the two new proposed MAC layer protocols for swarm operations
and presented some initial results. Assuming ideal channel conditions and using a linear single
dimension structure, an analytical model was described in this Chapter to evaluate network
delays and the impact that has on maximum number of vehicles that can be supported in a
swarm. Both the AST-TDMA and ATP-MAC protocols have exhibited substantial advantages
over the conventional TDMA protocol for the different applications that they have been designed
for. It has been shown that the new adaptive protocols outperform TDMA in their ability to
disseminate time-sensitive information in a timely manner and therefore allow much higher
densities of vehicles to operate in swarm like networks for both Cluster and Bus Topologies
studied.
The advantages of both the new adaptive protocols over TDMA and other time-based protocols
is that they can avoid the need for time synchronisation and significant guard times due to the
uncertainty that time synchronisation inflicts, which are both major drawbacks in time based
protocols. They also work with the spatial-temporal diversity created in long propagation delay
environments to allow ‘non-exclusive channel access' in a single channel, while maintaining the
collision avoidance benefit of contention free protocols. In addition a new metric, NCCP, has
been developed and presented to test the new MAC protocols for swarm operational
effectiveness.
This chapter has established the maximum possible swarm size for various coverage areas
based on the QoS requirements to ensure that the behavioural swarm formation algorithms can
be implemented via explicit communication between vehicles in the swarm. The network limit,
NCCP was established and used to define the number of vehicles possible to operate in a
particular topology or application area. The sensitivity to the Packet Arrival Rate in the MAC
layer queue was also recognised and that this will continue to be explored further in the
remainder of this work.
Chapter 6 will develop the swarm cluster topology focusing on the AST-TDMA protocol and will
explore modifications and improvements to the protocol for better performance when working in
a non-ideal acoustic channel. The implementation of a non-ideal acoustic channel into an event-
driven simulation program is done to reflect a more ‘realistic’ operation environment and is
based on the work in Chapter 3. Throughput and the new metric Channel Capacity Utilisation
will be introduced and used to evaluate protocol performance based on variations in transmitter
power and packet length.
127
Chapter 6 AST-TDMA Protocol Simulation Analysis and
Evaluation in Non-Ideal Underwater Environments
6.1 Introduction The objective of this chapter is to evaluate the performance of the Adaptive Space Time –
TDMA (AST-TDMA) in non-ideal underwater channel conditions, where ‘realistic’ dynamic
characteristics of interference and noise are implemented along with path loss so that packet
losses at the receiver will occur when channel conditions worsen. The chapter will also
introduce the event-driven simulation environment, of OpNet™ that will be used to implement
these dynamic characteristics to test the operation and performance of the new protocol.
In Chapter 5, the analytical network delay analysis was presented and a new metric, NCCP,
was developed to test the new MAC protocols for their ability to exchange the time-critical
navigational data between vehicles needed to maintain swarm synchronisation. This chapter will
use packet level simulations in OpNet™ to test the protocol to establish the performance
limitations of the protocol and therefore operational limitations of a group of autonomous AUVs
based on the limitations of the communication channel to exchange the required localisation
information to allow a group of AUV’s to operate in a swarm-like fashion.
In the first section of this chapter the channel model will be developed for OpNet and the
parameters that will be used in ideal and non-ideal channel conditions will be set out. Section
6.3 will validate the OpNet model by comparing it to the MATLAB results presented in Chapter 5
in ideal channel conditions. Section 6.4 will discuss and analyse modifications to the new MAC
protocol to build further improvements into the protocol for better network resource utilisation.
Additional performance metrics of Throughput and a new Channel Capacity Utilisation metric
will be introduced to help determine the protocol performance and this is presented in Section
6.5. Protocol performance in non-ideal channel conditions will be untaken in Section 6.6 with
discussion and analysis.
6.2 Simulation Model and Methodology With the objective of evaluating the protocol in non-ideal conditions, OpNet™ (Optimized
Network Engineering Tool) Modeler, which is an advanced network simulation tool, will be used
for these simulations. OpNet can evaluate the impact of channel conditions on packet
transmissions and receptions and therefore provides a means to establish more realistic
protocol performance results.
128
6.2.1 Modelling of an Acoustic Underwater Channel & Physical Layer in OpNet
OpNet does not have an acoustic underwater channel model and therefore modifications to the
physical layer (or pipeline stages) were made to match the characteristics of a short-range
acoustic channel that was established in Chapter 3.
In OpNet, the wireless communication channel is modelled as a pipeline with 14 computational
stages as illustrated in Figure 6.1; the first 6 stages modelling the Radio Transmitter side (Stage
0 – Stage 5) are implemented in the transmitter and the last 8 stages modelling the Radio
Receiver side (Stage 6 – Stage 13) are implemented in the receiver. As a packet is transferred
between a transmitter and receiver in OpNet it undergoes a series of computations modelling
various aspects of the link behaviour through this pipeline. For this work the channel
characteristics of a radio wave are replaced by models of an underwater acoustic pressure
wave. Table 6.1 provides the details of the 14 pipeline stages and highlights the stages that
have required modification to represent an underwater acoustic channel. The background to
these modifications can be found in Chapter 3.
The modification required on the transmitter side is for calculating the propagation delay on a
packet, which when mobile includes both the delay at the start of the packet reception and the
propagation delay at the end of the packet reception. The dra_sea_propdel stage obtains the
range between the transmitter and receiver at the beginning and end of reception and uses the
slower underwater acoustic signal propagation of 1500m/s, Section 3.3.3. This was set as a
constant in this research due to the insignificant impact that it was shown (in Chapter 3) to have
on short-range operations.
The modifications required on the receiver side include the received power and background
noise stages. Details of the modifications required can be found in Chapter 3, however an
overview of each of these is described here. The Receiver Power stage, which calculates the
path loss experienced by a signal has been adjusted to include a spherical spreading loss plus
Thorp’s attenuation loss model [132], Equation 3.7. The spreading loss is range dependent
while Thorp’s attenuation model is signal frequency and range dependent. The transmitter
signal level is converted to dB using the underwater reference pressure level Iref and an
electrical-acoustic efficiency conversion in the transmitter (projector) is estimated to be 50%.
The receiver (hydrophone) sensitivity is defined in terms of voltage response to pressure which
reflects the efficiency of the piezoelectric transducer and this acoustic-electrical efficiency will
also be estimate at 50% in this work as discussed in Section 3.2. The assumption taken is that
50% of either the applied electrical or mechanical energy is converted.
Table 6.1: Modified Pipeline Stages
Transmitter Receiver
Stages Name New file name Stages Name New file name 5 Propagation Delay dra_sea_propdel 7 Receiver Power dra_seathorp_power
9 Background Noise dra_sea_bkgnoise
129
Figure 6.1: Radio Transceiver Pipeline Execution for One Transmission
Noise at the receiver in underwater communication systems has unique and complex
characteristics that are frequency dependent. The background noise pipeline stage
dra_sea_bkgnoise has been modified to include the thermal noise of the transceiver and the
ambient noise of the ocean as defined by Urick [134]. These noises are calculated in terms of
the reference underwater pressure Iref. The ambient noise incorporates the noise associated
with the water movement around the hydrophone itself (turbulence noise), and the surface
motion noise associated with wind (of 1 m/s), wave and rain (wind noise). Shipping and other
very low frequency noise models have been included although they will have minimum impact
on the noise level as they lie outside the communication frequency range being used. Self-noise
associated with the vehicle itself has not been included because of its unlikely impact on signal
frequencies used. Intermittent noises of animals, ice cracking, and earthquakes that occur
within the communication frequency range are very difficult to accurately predict, as they will
130
vary from deployment to deployment. Various statistical models for intermittent noise sources
are available in the literature, however this is not a focus of this thesis and will not be
considered further.
6.2.2 OpNet Model Figure 6.2(a) illustrates a 5-vehicle Cluster Topology, V1 to V5, used to demonstrate the
performance of the AST-TDMA protocol. Figure 6.2(b) shows the OpNet Node model that
incorporates a Queue Process Module, AST_TDMA_1, shown in Figure 6.3. The process
model for V1 is slightly different to the other vehicles but only in its global variable assignment
and simulation control procedures that include the completion of the simulation at 10,000
packets. The MAC layer functionality is exactly the same in all vehicles based on the fully
distributed protocol design. The Process Module code is shown at Appendix B.
6.2.3 OpNet Parameters
The channel and transmission parameters used in OpNet simulations are summarised in Table
6.2 and reflect those used in the Matlab analysis of Chapter 5 (Table 5.7). Also, results
presented are those of vehicle V1, unless otherwise stated. Variations in results between
vehicles are generally small but do occur due to the vehicles position and sequence within the
cycle. Simulations were terminated when V1 has successfully received 10,000 packets.
The simulations are based on a swarm of AUV’s defined as a collection of V vehicles randomly
positioned in a Cluster Topology with ranges between consecutively transmitting vehicles being
similar. The swarm will be travelling at 4 knots (2 m/s), with each vehicle collecting its own
sensor data that has been defined in Chapter 5. Vehicles will use this sensor data to calculate
their own next trajectory based on a swarming formation algorithm that requires the location
data from all other vehicles in its neighbourhood.
Figure 6.2(a): 5-Vehicle Cluster Topology. Each vehicle represented
Figure 6.2(b): OpNet Node Model
131
Figure 6.3: OpNet AST-TDMA Process Model
The explicit communication mechanism between vehicles is through an underwater wireless
acoustic medium using a single frequency band for transmission. Each vehicle is capable of
sending and receiving data at a channel data rate of 9600 bps. The transceiver will operate in
half-duplex mode, that is, it will be either transmitting or receiving but not both at the same time
and a single omni-directional antenna is mounted on each small vehicle.
Simulations will be based on a stable structure for the comparison with the analytical results
from Chapter 5. The stable structure means that all vehicles are moving in the same direction at
the same speed and that all vehicles will maintain the same range between each other during
the simulations. Mobility was therefore not considered as relative velocity was zero.
6.3 Validation of Simulation Model To confirm the operation of the OpNet simulation model, the simulation results in this section,
are based on ideal channel conditions and evaluated and compared with the analytical results
of Chapter 5. That is the transmission power level is set to ensure a high SNIR at the receiver
so that there is no BER when signals are received and therefore can guaranteed a 100% packet
success rate.
132
Table 6.2: Main Parameters and Transmission Characteristics used in OpNet
Channel Transmission Acoustic Signal Speed 1500 m/s Swarm Size 5 - 40 vehicles
Centre Frequency 40 kHz Average Range between V 20 – 50 m
Bandwidth 10 kHz Transmission Rate 9600 bps
Attenuation Model Thorp [132] Data Packet Size From Table 5.2
Wind 1 m/s Transmission Power 1 – 3 W
Channel Reverberation L1 Low Sea: -54 dB ± 3 dB Transducer efficiency 50%
Channel Reverberation L2 High Sea: -46 dB ± 3 dB Modulation BPSK
Swarm Reverberation Assume Perfect reflection BER 10-4
Swarm Reverberation With anti-phase signals Minimum SNIR 8.6 dB
6.3.1 Protocol Process Evaluation A visualisation of the protocol in operation and the packet transmission and reception process
has been presented in Figure 6.4 and with a comparison to the TDMA protocol presented in
Figure 6.5 for a 5-vehicle Swarm Topology, V1 to V5 shown in Figure 6.2 (a). In this example the
OpNet simulation has been stopped after V1 received its 20th packet. Each packet is 100 ms
long (using 9600 bps & 960 bits) represented as the solid bars and the ‘spaces’ represent the
time when the vehicle is idle, that is, not transmitting or receiving a packet. The red bar in the
bottom graph of Figure 6.4 illustrates the first packet transmission from V1. That packet is then
received by V2 to V5 (blue bars) at increasing delays, as the first activity in each of these
vehicles. When V2 receives the packet, it gets the trigger to transmit, which it does immediately.
V2’s packet is then received by V1, V3, V4 & V5, with V3 being the next to transmit.
Figure 6.5 shows only the operation of vehicle V5, where the top graph of Figure 6.5 is the same
as the top graph of Figure 6.4. Figure 6.5 demonstrates the use of the TDMA protocol and
shows the timing comparisons between the AST-TDMA and TDMA protocols. This comparison
illustrates the reduction in NCCP time when using AST-TDMA compared to a conventional
TDMA protocol. Five full cycles are completed using AST-TDMA protocol before the fourth cycle
is completed using the TDMA protocol. These preliminary discussions on these time differences
were begun in Chapter 5 and will be explored in more detail in this chapter in realistic channel
conditions.
As a broadcast medium, all the vehicles in range will receive the sent message that will be
received at times based on propagation delays. For the TDMA protocol, the slot-duration is
based on the time it takes for the last vehicle to complete reception, and therefore all receptions
will occur within a slot used to send out the packet, thus maintaining exclusive access in each
slot period. Figure 6.4, however, shows that the AST-TDMA protocol can allow non-exclusive
access by allowing vehicles next in sequence to start transmitting before all vehicles have
received the previous packet without causing collisions. This is demonstrated in Figure 6.4 with
the vertical lines starting at ~3.7 s that show one NCCP cycle. At the start of the cycle, V1
(bottom graph in Figure 6.4) transmits and the other vehicles begin to receive V1’s packet as it
133
propagates towards V5. The second vertical line shows the start of V2’s transmission, which
crosses V1’s packet still being received by V3, V4 & V5. This shows that the AST-TDMA protocol
utilises non-exclusive access of the channel while avoiding collisions, which is how it can
improve its channel utilisation.
Figure 6.4: AST-TDMA Protocol, 5-Vehicle Swarm of Figure 6.1(a), illustrating Packet Tx & Rx in each vehicle
Figure 6.5: Comparison of Timing between AST-TDMA & TDMA, for V5 of 5
134
The operations of the AST-TDMA protocol can encounter periods with large idle times, note the
period following V5’s transmission of it’s packet top graph in Figure 6.5, when sequenced
vehicles are located at ranges where the propagation delay is much larger than the
transmission times. This is because V5 has to wait while its packet propagates to V1 and then for
V1’s packet to return, which means 2 x the maximum propagation time in this scenario. This
time is the same as for V5 in the TDMA protocol, see bottom graph in Figure 6.5, as the TDMA
slot size is based on maximum propagation time. In this simulation, the TDMA slot-time did not
include an additional guard-time to take into account mobility and the potential changes in
range, thus, the timings shown for the TDMA results are best case and used as a comparison
with the AST-TDMA protocol. Thus, it illustrate that the AST-TDMA protocol can have a slot size
approaching that of TDMA depending on the positioning of the vehicles within a Topology and
the scheduling of vehicles within a cycle. A guard time of maximum propagation delay is
incorporated in all other analysis of the TDMA protocol.
In addition, Figure 6.4 shows that V4 does not have a packet to send in the third round. It instead
sends a token (not shown) that allows the cycle to continue. This will mean that the other
vehicles will need to wait until the second cycle to receive an updated packet from V4, which
increases the NCCP. Thus, it is important to ensure that there is a packet ready to send,
however also to ensure that the packet is not generate to early, as the waiting time in a vehicles
queues will add to the age of the data that will be used by the swarm formation control
algorithm. This can increase the true NCCP.
(a) 15 Vehicle (b) 5 Vehicle
Figure 6.6: 5 and 15 Vehicle Cluster Topology Swarm, @ 50 m, initial positions
135
6.3.2 Validation of Simulation Model Results With the design and set-up of the protocol in the OpNet environment, this section will compare
the results with those obtained from the analytical work completed in Chapter 5 to validate the
OpNet model. Figure 6.7 compares the 5 and 15-vehicle swarms of Figure 6.6 (a) and (b) using
the results of V1, with the results presented in Figure 5.11. It can be seen that there is no
discernable difference between the results obtained from the simulations done in OpNet
compared with those produced in MATLAB. The matching up of results particular for the TDMA
protocol is explained as in ideal conditions with no packet loss, the cycle time is depended on
the fixed slot length and number of vehicles in the swarm, which is accurately simulated in
OpNet. For the AST_TDMA the Tcycle is dependent on the range between sequenced vehicles
that is set at 50 m in MATLAB, however dependent on placement of vehicles within the
structure of the OpNet model. In the 15-vehicle swarm of Figure 6.6(a) the actual average rss is
slightly longer at 51.4 m which explains the fractionally larger Tcycle of this swarm.
Figure 6.7: Comparison of Cycle Time (Tcycle) obtained in OpNet and MATLAB for both the AST-TDMA and TDMA protocols (compare with Figure 5.11)
0
0.5
1
1.5
2
2.5
0 0.5 1 1.5 2 2.5 3
Cycl
e Ti
me:
Tcy
cle
(s)
Packet Arrival Rate per Vehicle (pkts/s)
15-Vehicle TDMA MATLAB 15-Vehicle OpNet 15-Vehicle MATLAB 5-Vehicle TDMA MATLAB 5-Vehicle OpNet 5-Vehicles MATLAB
Arrival Rate Arrival Rate when
136
Figure 6.8: AST-TDMA protocol showing relationship between Tcycle, NCCP and Packet discard. Comparison of OpNet and MATLAB results (compare with Figure 5.12)
Figure 6.9: TDMA protocol showing relationship between Tcycle, NCCP and Packet discard. Comparison of OpNet and MATLAB results (compare with Figure 5.14)
0
0.5
1
1.5
2
2.5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Cycl
e Ti
me:
T
cycl
e (s
) and
NCC
P (s
)
Packet Arrival Rate per Vehicle (pkts/s)
NCCP (s) OpNet NCCP (s) MATLAB Tcycle OpNet/MATLAB Queue Discards (pkts) OpNet Queue Discards (pkts) MATLAB
Num
ber o
f Pac
kets
(pkt
s)
1
0
2
A packet discarded
Cycle Saturation
0
0.5
1
1.5
2
2.5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Cycl
e Ti
me:
Tcy
cle
(s) a
nd N
CCP
(s)
Packet Arrival Rate per Vehicle (pkts/s)
NCCP (s) OpNet NCCP (s) MATLAB Tcycle (s) OpNet / MATLAB Queue Discards (pkts) OpNet Queue Discards (pkts) MATLAB
A packet discarded
Num
ber o
f Pac
kets
(pkt
s) 2
1
0
Cycle Saturation
137
Figures 6.8 and 6.9 compare the OpNet results of only the 5-vehicle swarm of Figure 6.6 (a),
with the MATLAB results presented in Figures 5.12 and 5.14 for the AST-TDMA and TDMA
protocols respectively. The only real difference between the MATLAB and OpNet results are
seen in the NCCP values at various Packet Arrival Rates below cycle saturation where the
OpNet values are not decreasing at a linear rate. This is due to the fact that OpNet can evaluate
the actual NCCP for a particular vehicle, that is, it can determine the starting time of a packet
transmission from that vehicle to the time when that vehicle receives all the packets, rather than
for MATLAB that calculates an average value. The zigzag is also partly affected by the
increments of Packet Arrival Rate intervals used for the calculations of NCCP at 0.1 pkt/s (in
MATLAB 0.001 pkts/s was used).
6.4 Protocol Modifications This section will investigate modification to the basic AST-TDMA protocol used so far to gain
further improvements in network resource utilisation and the performance of an USSN. From the
MAC layer perspective, the timing of the arrival rate of the sensor and navigational data from
the application layer, which cannot be controlled at the MAC level, has a significant impact on
the operations of the protocol.
This section will examine the important procedural aspects of the protocol for effective
operations and considered modifications that can be incorporated in the protocol to improve the
network resource utilisation. Two additional metrics to evaluate the network will also be
considered in this section.
6.4.1 Protocol Procedures Although the AST-TDMA protocol is an adaptive time based protocol there are still several time-
critical elements of the procedures within the protocol that impact on its effectiveness to
exchange data within the swarm to maintain swarm synchronisation and thus the QoS boundary
of NCCPlimit that was introduced in Chapter 5.
For the AST-TDMA protocol to succeed in real time, there are four important procedural matters
that need to be considered, which are illustrated in Figure 6.10.
1. Matter 1, in Figure 6.10, is to ensure that packets do not collide despite concurrent
transmissions being used due to the temporal-spatial diversity experienced underwater, and
this was discussed and confirmed in Section 5.5.
2. The determination of an optimum schedule, Matter 2, as it will impact on the overall cycle
time, as slot lengths will change with variations in ranges between sequenced vehicles. This
aspect will not be considered further in this work, however there have been several recent
publications investigating scheduling more generally in underwater communication research
[61, 81, 79]. The schedule arrangement used in this work is based on vehicle ID and LIDCA
and the placement of vehicles in the swarm at the beginning of each simulation, which is
done to minimise sequenced vehicle ranges.
138
Figure 6.10: AST-TDMA considerations in protocol operations
3. Matter 3 relates to the dependency on the Packet Arrival Rate (λ) in each vehicle’s MAC
queue to the protocol functionality and performance. The current protocol is designed
around a deterministic packet arrival rate (λ) of sensor data packets being delivered from
the application layer to the MAC queue in each vehicle. There are two modes of operation
that that have different outcomes that are based on network operations above or below
cycle saturation.
Matter 3 (a): If Packet Arrival Rate is too low (λ < λmin and M > 1) (below cycle saturation)
then some vehicles may not have a Data Packet to send which creates a problem for the
protocol that relies on a trigger from sequence of back-to-back propagation of packet. To
ensure the cycle is not broken a small Token Packet is sent from the vehicles that do not
have data to send, which carries the trigger instead to the next vehicle. The impact of this
solution is that several cycles will be needed for full exchange of Data Packets from all other
vehicles, which increases NCCP. The second consequence is that some packets that were
successfully received in a vehicle may be destroyed due to the arrival of a newer packet
from the same transmitting vehicle in a previous cycle but within a NCCP exchange period.
The modification to the protocol that is introduced to deal with this scenario is to add a
‘WAIT’ time to the vehicle that has the token. This will be explored in Section 6.5.
139
Matter 3 (b): If, on the other hand, Packet Arrival Rate is too high (λ > λmin and M = 1), then
the MAC queue will grow, and the queue will be flushed of the oldest packets in the LIFO
queue so that only the youngest packet (and information) is sent. That is, packets will be
discarded from the queue and therefore lost from circulation. The ability of the protocol to
handle larger packet sizes will be investigated as a modification to allow the incorporation of
payload data and data packet trains if there is that information waiting in the queue and the
network is within QoS boundaries. This will be explored in Section 6.5.
4. The final Matter, Matters 4 as shown in Figure 6.10 relates to the loss or error of a packet
that is associated with a non-ideal channel. This matter deals with the reliance on the
reception of packets in each sequenced vehicle, as the AST-TDMA protocol requires back-
to-back transmissions to ensure that the cycle is not disrupted. To deal with this scenario,
various back off options were investigated and these presented in Section 6.6.
In addition, for non-sequenced vehicles this would mean that NCCP would increases as the
vehicle would have to wait another cycle before it could gain information from the vehicle who’s
transmission that did not arrive. The option of ‘piggybacking’ data from the previously received
packet is investigated. It’s impact on the performance of the protocol is the same as requiring a
larger packet size to be incorporated, which is discussed in Section 6.5.
6.4.2 Additional Protocol Performance Metrics The analysis completed in Chapter 5 focused on the timing delays experienced by data in a
swarm network and the boundaries imposed by the requirements to ensure exchange of
localisation data to guarantee swarm synchronisation and the new metric NCCP was
developed. NCCPlimits were established based on the Quality of Service (QoS) requirements of
the application and from this the maximum number of vehicles that the protocol can support was
determined.
Once the network is operating within the QoS boundary, the evaluation and improvement in the
utilisation of the network resources is an important objective of the protocol performance.
Therefore, the metrics of Throughput and Channel Capacity Utilisation and the Information
Convergence Time of data in one vehicle will be analysed to determine performance of the
different protocol extensions.
6.4.2.1 Throughput Throughput, G, is defined here as a measure of successfully received packets from those
offered to a vehicle, that is, the rate at which traffic is successfully received at a vehicle. The
normalized throughput is the number of successfully transmitted bits over a fixed period divided
by the channel capacity (data rate R). The normalised throughput for the ith vehicle, Gi is:
𝐺𝑖 = ∑ (𝑃𝑖 𝐿data)𝑖𝑅∗𝑇sim
(6.1)
where Pi is the number of packets successfully received at i, Ldata is Data Packet length (bits),
and Tsim refers to the time over which packet count is taken.
140
(a) Normalised Throughput (b) Channel Capacity Utilisation
Figure 6.11: Comparisons of Protocols and Number of Vehicles in Swarm at 50 m
It can be seen in Figure 6.11 (a) that the AST-TDMA protocol provides a better throughput than
for the TDMA protocol at around twice as many packets above cycle saturation. Once cycle
saturation occurs the throughput becomes constant as expected with constant ranges between
sequenced vehicles as the cycle time and number of packets being received remains constant.
That is, once saturation of the cycle has occurred, which is the successful reception of a packet
in each vehicle, throughput hits its maximum respective of packet arrival rate as the extra
packets generated between cycles will be discarded.
The throughput for the AST-TDMA protocol at Packet Arrival Rates lower than cycle saturation
varies considerable due to the sensitivity to the timing of packet arrivals and the Tcycle. With a
small change in arrival rate, a packet may or may not be ready to send and thus this jagged
response occurs depending if a vehicle has just obtained a packet in that cycle or if several
cycles are needed. This ‘saw-tooth’ effect gets considerable more dramatic as the number of
vehicles in the swarm reduces, as the cycle times become shorter and there are less slots in a
cycle. Therefore the difference seen between the 15 and 5 vehicle swarm throughput values are
due predominately to the percentage of time (or slot) used for a transmission of a packet
compared with the number of slots used for reception which can also be seen in the Channel
Utilisation value.
6.4.2.2 Channel Capacity Utilisation Where throughput focuses on successfully received packets, Channel Capacity Utilisation
evaluates how busy the channel is or how well the protocol is able to use the limited channel
underwater based on channel conditions and utilising a single channel for all transmissions and
receptions of packets to all vehicles in the swarm without packet collision. That is, Channel
Capacity Utilsation is about how much of the time the channel is busy. In non-contention based
scenarios and ideal conditions, throughput and channel capacity utilisation are similar.
0 0.05
0.1 0.15
0.2 0.25
0.3 0.35
0.4 0.45
0.5
0 0.5 1 1.5 2 2.5 3
Nor
mal
ised
Thr
ough
put
Packet Arrival Rate per Vehicle (pkts/s)
15-V AST-TDMA 5-V AST-TDMA 15-V TDMA 5-V TDMA
0
10
20
30
40
50
60
0 0.5 1 1.5 2 2.5 3
Chan
nel U
tilis
atio
n (%
)
Packet Arrival Rate per Vehicle (pkts/s)
15-V AST-TDMA 5-V AST-TDMA 15-V TDMA 5-V TDMA
141
However, when the channel becomes contention based, throughput establishes the number of
successful transmissions compared with channel capacity utilisation which defines how much
time the channel is busy for.
For short-range USSN, there are no other uses that need to share the channel and as it is
limited the focus is on maximising channel utilisation as what is not used is lost. Thus the metric
that we wish to measure is how much of the channel capacity is used. Channel Capacity
Utilisation (Ui) therefore provides a perspective on the effectiveness of the protocol to utilise the
available channel capacity and establishes how much wasted channel time there is based on
the half-duplex, single channel operations over a fixed time period. This is based on one vehicle
in the swarm’s transmissions and successfully received Data Packets that will be useable in the
vehicles swarm formation algorithm.
In particular, we are comparing over a specified time period the amount of time used in
transmission and successful reception of packets (Titx, Ti
rx) in one vehicle including packet
overheads, to the total time period including idle time (Tidle), wasted time (Twaste) due to packets
unsuccessful delivered and guard time (Tguard) for processing and reverberation times. Thus the
Channel Capacity Utilisation, Ui is:
𝑈𝑖 % = ∑�𝑇𝑖𝑡𝑥+𝑇𝑖
𝑟𝑥�∑�𝑇𝑖
𝑡𝑥+𝑇𝑖𝑟𝑥� + 𝑇𝑖𝑑𝑙𝑒+ 𝑇𝑤𝑎𝑠𝑡𝑒+ 𝑇𝑔𝑢𝑎𝑟𝑑
* 100 (6.2)
Figure 6.11 (b) presents the Channel Capacity Utilisation difference between the AST-TDMA
and the TDMA protocol and a 15 and 5 vehicle swarm in ideal conditions. As expected with the
better throughput, the AST-TDMA protocol shows a much higher channel utilisation. The low
utilisation of the available channel for TDMA is due to the requirements for large guard times,
which force regular periods of time that cannot be used.
The values were based on the reception of 10,000 data packets in V1. Taking a Packet Arrival
Rate of 1.2 pkt/s the number of packet sent from V1 was 700. This occurred in 15 m 40 s for the
AST-TDMA protocol and 31 m 50 s for the TDMA protocol, which gives the Channel Capacity
Utilisation of 50.8 % and 24.9 % respectively, as demonstrated in Figure 6.11 (b).
6.5 Analysis of Protocol Modifications The two major protocol variations proposed in Section 6.4 are:
• the ‘Wait’ modification that focuses on improvement to the operations of the protocol below
cycle saturation and
• the variation to packet size required to support packet ‘Trains’ or ‘Piggybacking’ when the
network is operating above saturation and still within the QoS boundaries.
6.5.1 Transmission ‘WAIT’ Modification The WAIT modification is a replacement for sending a Token Packet when a vehicle does not
have a Data Packet in its queue to send. Thus, when a vehicle receives a token to trigger it to
send a packet it waits for the Arrival of a Data Packet in its queue from the application layer
142
which it will then access and send immediately. The OpNet process model is shown in Figure
6.12 and the model code is shown at Appendix E. This modification has no impact on the
protocol when it is operating above cycle saturation, λmin as there will always be a packet in the
queue to send.
There are hybrid versions to this modification where the wait time may be determined by other
parameters apart from the arrival of a Data Packet in the queue. These could include a wait
time based on maintaining knowledge of previous NCCP times and using an average or
maximum value of this to wait for a packet to arrive in the queue or more simply a
predetermined slot time similar to a TDMA time slot. Alternative, at the start of a wait time, the
vehicle could trigger the generation of a Data Packet in the application layer, thus pausing the
cycle until the vehicle creates a new Data Packet in a higher layer. These are not explored
further here but are suggestions for further work.
There are several advantages to the WAIT modification. If all vehicles have the same or similar
Packet Arrival Rate the wait time only needs to happen at one vehicle per cycle as the time that
it has to wait in that vehicle allows all other vehicles to have a packet in their queue. In addition,
when the swarm is moving and the range between sequenced vehicles are dynamically
changing so that Tcycle changes then the cycle saturation point will change. If vehicle Arrival
Rate is operating close to λmin the Wait modification can prevent regular additional cycles per
NCCP and therefore better ability to maintain performance with changes in topology.
The result of a wait time is that it forces only one cycle to occur, C=1, and therefore from
Equation 5.2, cycle time Tcycle equals NCCP time. Thus, Tcycle will increase to allow there to
always be a Data Packet sent in each vehicle each cycle. The major advantages to this
modification are that it avoids the use of a Token Packet that may itself get lost or be in error
which will take channel capacity away from sending useful data. It also eliminates the
requirement of additional cycles and retransmissions from vehicles who’s packet has already
been distributed.
Disadvantages to using the ‘Wait’ option may be seen when Packet Arrival Rates are very
different in each vehicle, as sending a token will allow the rest of the swarm to keep exchanging
packets while that vehicle without a packet has the time until the next cycle for a packet to arrive
in it’s queue. This is an unlikely scenario for swarming networks but is an interesting problem for
future work.
Figure 6.13 compares the difference between the original AST-TDMA with Tokens and the
TDMA protocol with the proposed modification to AST-TDMA with WAIT by showing the number
of cycles required per NCCP. The Packet Arrival Rate at which cycle saturation occurs is 0.77
pkts/s for the AST-TDMA protocols and 0.38 pkts/s for the TDMA protocol. For the TDMA
protocol, the λmin is much lower and therefore the number of cycles required per NCCP is lower,
as Tcycle is higher at each λ due to the longer slot times required for guard times.
143
Figure 6.12: Process Model for Wait Modification
The AST-TDMA with Wait protocol only requires one cycle per NCCP above and below λmin
while the AST-TDMA with Tokens, as has been seen in Chapter 5, will require many cycles to
ensure the exchange of all vehicles information below λmin or until there is a packet in each
vehicle each cycle and therefore only one cycle is required for NCCP. The dramatic variation in
number of cycles per NCCP for the AST-TDMA protocol between adjacent Packet Arrival Rates
demonstrates the impact that the timing of packet arrivals, cycle time and NCCP time can have
as discussed in Section 6.4.1. This miss alignment of timing can also be seen in the TDMA
protocol when the time of a packet arrival has just missed a cycle but due to the large slot times
this occurs less regularly. The Wait mechanism has an advantage of smoothing out the
variations as Packet Arrival Rates change and this will be seen to be particularly useful when
operating close to cycle saturation.
Figure 6.14 shows the comparison of NCCP for the three protocols and the Tcycle for AST-TDMA
with Token (shown previously in Figure 5.10). The Tcycle for AST-TDMA with Wait is equal to its
NCCP as the wait mechanism forces a single cycle per NCCP and the Tcycle for the TDMA
protocol is constant based on the number of vehicles and the slot-time (these are not shown).
The clear benefit of AST-TDMA with Wait protocol is demonstrated by the reduced NCCP time
for all values of λ below λmin and is equal to (at low λ) or better than the NCCP for the TDMA
protocol.
144
Figure 6.13: Comparison of Average Number of Cycles per NCCP
Figure 6.14: Comparison of NCCP times between protocols and Tcycle for the AST-TDMA with Token
Figure 6.15: Comparison of the true Channel Utilisation Uitrue
0
1
2
3
4
5
6
7
8
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 Aver
age
Num
ber o
f Cyc
les
per
NCC
P (
C)
Packet Arrival Rate per Vehicle (pkts/s)
AST-TDMA with Token TDMA AST-TDMA with Wait
0
1
2
3
4
5
6
7
8
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 Tcyc
le (s
)
N
CCP
(s)
Packet Arrival Rate per Vehicle (pkts/s)
AST-TDMA with Token AST-TDMA with Wait TDMA
0 5
10 15 20 25 30 35 40 45 50 55
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85
Chan
nel U
tilis
atio
n (%
)
Packet Arrival Rate per Vehicle (pkt/s)
AST-TDMA with Wait AST-TDMA with Token TDMA
145
The advantage of the reduced in NCCP can also be seen in the Channel Capacity Utilisation
shown in Figure 6.15, where the adaptive protocols show significant higher utilisation of the
channel resources comapred with the conventional TDMA protocol when operating above the
TDMA protocols cycle saturation. It is at this λsat that the maximum channel utilisation is
achieved which is only 25% for TDMA and just over 50% for the AST-TDMA protocol. These
values reflect the reduction in channel resources that the TDMA protocol can obtain due to the
large guard times required in each slot when operating on or over cycle saturation point.
The similar channel capacity utilisation seen when the network is operating below cycle
saturation is due to the similar NCCP times that both the AST-TDMA with Wait and the TDMA
protocol can achieve. Thus, if operating in this region, the TDMA protocol is an alternative
option to the AST-TDMA with Wait.
When comparing between the two modifications of the adaptive protocol, the AST-TDMA with
Wait protocol is consistently higher than the AST-TDMA with Token due to the reduction in
unnecessarily cycles that is achieved with the wait modification. And again that ‘saw-tooth’
result for AST-TDMA with Token occurs when the timing between packet arrivals in a vehicles
queue and when it has permission to access the channel requires extra cycles.
Therefore to operate at maximum channel resource utilisation the network operation needs to
be on or above cycle saturation. As the operating environment is dynamic and operations are
requiring a λ ~ λsat then the AST-TDMA with Wait protocol is superior to the AST-TDMA with
Token due to the dramatic reduction in Channel Utilisation just below λsat and provides a method
to smooth out the fluctuations in the number of cycles per NCCP, NCCP and network resource
use.
Thus, the Wait mechanism allows more flexibility with variations in cycle times and packet
arrival rates and improves channel capacity utilisation by reducing the number of packets
unnecessarily transmitted either as tokens or those that are destroyed due to a newer packet
being received within the NCCP.
6.5.2 Packet Size Variations There are two aspect of the operation of the adaptive protocol where the need to increase
packet size could provide benefits to the swarm operation as discussed in Section 6.4. The first
of these is when the network is operating above saturation and packet information is being lost
from being exchanged between vehicles in the swarm because to the current requirement that
only the most up-to-date locational data packets are used. An option of a Data Packet Train
could allow more information to be exchanged, but this would only be possible if the network
was operating below the QoS boundary. The second suggested potential benefit of increasing
packet size is to provide a mechanism to reduce cycle times in the case when a packet does
not arrive at a vehicle due to channel errors. Here an option of “Piggybacking” Data Packets on
top of the Data Packet that is sent from each vehicle is proposed.
146
Figure 6.16: Packet Discards per cycle: Comparison between protocols for a 15 and 5 vehicle swarm at 50 m
6.5.2.1 Option of Data Packet Train Once cycle saturation has been reached, then a vehicles queue will begin to grow. When this
occurs, the number of packets that are discarded from the MAC queue becomes a measure of
the potential information lost from exchange around the swarm and the ability of the protocols to
service these packets in an underwater environment.
The number of packets that will be discarded in a vehicle per cycle based on and increasing
Packet Arrival Rate is illustrated in Figure 6.16. This figure shows that the TDMA protocol, for
the same number of vehicles in a swarm, is not able to service the same number of packets as
the AST-TDMA protocols, as the longer cycle times mean that there is less opportunity to send
a packet. This further reinforces the better performance of the AST-TDMA protocol compared to
the TDMA protocol as shown in the Throughput and Channel Utilisation results of Figure 6.12.
At low Packet Arrival Rates, up to cycle saturation, there will be no packets discarded, as there
is not always a packet in the queue to send. This occurs at a lower Packet Arrival Rate for
TDMA due to the higher cycle times required by this protocol. As discussed in Section 5.9.4,
above cycle saturation, it will require several cycles before a packet is discarded. The number of
cycles between discards will decrease until it reaches a Packet Arrival Rate when there will be a
packet discarded each cycle That is when the queue receives 2 packets during each cycle
period. For the 15-Vehicle swarms, a packet will be discarded each cycle at 0.7 pkts/s for TDMA
and 1.5 pkts/s for the AST-TDMA.
If the network is not operating close to the QoS boundaries and there is time in the network
cycle for additional information could be sent, the option of a packet train would allow the
exchange of larger data packets. The packet train arrangement would introduced the ability to
0 1 2 3 4 5 6 7 8 9
10
0 0.5 1 1.5 2 2.5 3 3.5 4
Num
ber
of P
acke
ts D
isca
rded
per
Cy
cle
per
Vehi
cle
(pkt
s)
Packet Arrival Rate per Vehicle (pkts/s)
15-V TDMA 15-V AST-TDMA 5-V TDMA 5-V AST-TDMA
147
not only send the youngest packet in the queue (as implemented by the LIFO queue) but
information from previously queued packet could also be sent which would allow a larger
amount of information to be transferred. This is equivalent to requiring an increase in packet
size, which would only be required above cycle saturation.
The impact of increasing Data Packet sizes on NCCP for the AST-TDMA with Wait and the
TDMA protocol shown in Figure 6.17 (a) and (b) respectively. In both protocols, as packet size
increases so does the NCCP at and above saturation as expected due to the higher cycle times
that will be required due to the higher transmission times and therefore slot lengths. Examining
the comparison between the 424 and 1224 bit packets, there is a difference in transmission time
of 0.044 s to 0.127 s per slot (𝐿𝑑𝑎𝑡𝑎𝑅 ) respectively, which is a total of 1.25 s for a complete 15-
vehicle cycle. This is the variation seen in Figures 6.17 (a). For the TDMA, Figure 6.17 (b), it is
the difference in slot time by the number of slots required per cycle. For the 15-vehicle swarm
shown the difference seen in steady state NCCP for the 424 bit and 1224 bit packet is 1.25 s
which is the difference in the slot times of the different size packets or 0.1775 s and 0.261 s
respectively as determined by Equation 5.4.
While an increase in packet size may provide a means of exchanging more information, there
will be no benefit if the network does not remain within the application QoS boundary NCCPlimit.
Taking the QoS boundary determined in Chapter 5, of a maximum of 3 s the AST-TDMA with
Wait protocol, under ideal conditions, can operated with larger packets as shown and still be
within the boundary when operating on or above cycle saturation. For the TDMA protocol only
the smallest size packet is possible, as the Data Packet Sizes above 664 bits (NCCP = 3.0013
s), is over the NCCPlimit. An option of a smaller swarm size would need to be considered for the
TDMA protocol to operate successful.
(a) NCCP for AST-TDMA with Wait (b) NCCP for TDMA
Figure 6.17: Comparison of NCCP for 15-vehicle swarm at 50 m with various Data Packet Sizes defined in Table 5.9
0 0.5
1 1.5
2 2.5
3 3.5
4 4.5
5
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
NCC
P (s
)
Packet Arrival Rate per Vehicle (pkts/s)
1224 bits 744 bits 664 bits 424 bits
0 0.5
1 1.5
2 2.5
3 3.5
4 4.5
5
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
NCC
P (s
)
Packet Arrival Rate per Vehicle (pkts/s)
1224 bits 744 bits 664 bits 424 bits
148
Figure 6.18: Channel Capacity Utilisation at λsat: Comparing AST-TDMA with Wait and TDMA protocols
Figure 6.18 illustrates that both protocols benefit from with increasing channel utilisation as
packet size increases, which is due to the higher percentage of time that is used for data
transmission to propagation delay. However, as just discussed, while the packet size limits
increases the NCCP beyond its boundary, the increase in channel resources cannot be used.
6.5.2.2 Option of Piggybacking of Data Packets The option of ‘piggybacking’ data packets is suggested for supporting improved information
exchange when packet losses due to channel error will force additional cycles to obtain full data
exchange within the swarm. That is, if V3 for example has successfully received V2’s packet so
that it is now V3’s turn to send a packet, then V3 can include V2’s information in its packet for
other vehicles to obtain V2 information as a redundancy measure that would reduce NCCP. This
mechanism is suggested for further work however an examination of the impact of increasing
packet size has on a swarm will be discussed further in Section 6.6.
6.6 Results in Non-Ideal Underwater Environments 6.6.1 The Non-Ideal Channel in OpNet The non-ideal channel was simulated with variations in the Noise model (dra_sea_bkgnoise)
that include changes in Reverberation Level (RL) and the Path Loss model
(dra_seathorp_power) that include changes in transmitter power. The reverberation levels used
here are those discussed in Section 3.3.4 and represent the reverberation energy that decays
following the reception of sound energy including volume scattering. As shown in Waite [137]
0
10
20
30
40
50
60
70
80
424 664 744 1224
Chan
nel C
apac
ity
Uti
lisat
ion
(%)
Data Packet Size (bits)
AST-TDMA with Wait TDMA
149
the reverberation levels for high sea state, when the sea is turbulent and rough, is
approximately RL= -46 dB while at a low sea state, or very calm conditions, the RL = -54 dB.
To illustrate the reception of packets in a vehicle, Figure 6.19 provides a snapshot of a time
period during a simulation for a 5-Vehicle swarm operating at 20 m using a transmitter power of
1 W. The variations is SNIR that vehicle V1 sees for each of the 4 packets that it must receiver
per cycle, can be seen in Figure 6.19, in low sea state (top graph) and high sea state (bottom
graph). The AST-TDMA protocol is used here with an arrival rate in each vehicle of 1 pkt/s
which is the reason for the groupings of 4 packets (represented as dots) that occur every
second due to the inter-packet arrival rate of 1 s. The variations in each packets’ SNIR level as
it is received by V1 is based on the variation in reverberation level at the time each packet is
received, that includes the presence of the other vehicles operating in its vicinity and the range
(which relates to attenuation loss) that a packet coming from each of the other 4 vehicles will
experience. The timing of packets being received by V1 is based on range and the MAC
protocol slot timing, which in this case is operating at a λ < λsat.
Figure 6.19: SNIR for 5-Vehicle Swarm at 20 m
150
At low sea state and Ptx = 1 W provides close to ideal conditions with an average SNIR of
higher than 15 dB. In this case, depending on projector sensitivity a lower transmission power
could be used. However, as the sea state changes to a high level, the reverberation level
increases and the SNIR reduces with some packets falling below the minimum SNIR set for
successful reception (8.6 dB). These packets will be rejected in the simulation and considered
undelivered in that vehicle meaning that it will need to wait a cycle before it has another chance
of receiving a new packet from the vehicle who’s packet was in error.
A BER threshold is set at 10-4 which is used to compare with the actual BER in a receiving
packet that is determined from the BPSK modulation and SNIR of the receiving signal. If the
actual BER is lower than this threshold the packet is accepted and it is assumed that the bit
errors are corrected by an error correction mechanism. If the BER is higher than this threshold
then the packet will be rejected.
The remainder of this chapter will examine various scenarios under non-ideal conditions.
6.6.2 Variations in Transmitter Power A 5-vehicle swarm, as shown in Figure 6.6 (b) that is operating at an average range between
vehicles of 50 m is used to compare the results of variations in transmitter power. Figure 6.20
illustrates the average Packet Loss that will be experienced due to the average SNIR that
occurs on all the packets received by V1 for a transmitter power of 1 W and 3 W. For each of
the transmitter power levels, the SNIR level is governed here by the reverberation level that can
be seen as a reflection of the sea state changes shown in the figure. The groups of points
shown for each power level and SNIR, reveal the small variations that occur based on a series
of simulation at different Packet Arrival Rates and reverberation level dynamics.
As expected, at both transmitter power levels, the packet losses are increasing with decreasing
SNIR. When the swarm is operating with a higher transmitter power, the results in red in Figure
6.20, the level of reverberation that can be dealt with is much higher. Take for example the
furthest left hand red group of values, in Figure 6.20, that represent results using a Ptx = 3 W
and show a packet loss of ~22% for a very high sea state (or reverberation level of - 45 dB).
The blue group next to this red group, that has a lower transmitter power of 1 W, has a similarly
large packet loss of ~ 19% in a much lower sea state (or reverberation level RL of -50 dB). This
relationship continues down the packet loss curve where the network operating with a higher
transmitter power can achieve almost zero packet losses even with a mid to low reverberation
level. It can also be seen that at approximately the same SNIR the packet losses are similar
irrespective of transmitter power, which is expected. What these results clearly show, however
is the requirement to increase transmitter power in higher sea states, as is predictable.
Recalling Equation 3.24, and simplifying it to illustrate this relationship where SNIR is based on
the received signal level and compared to the combination of all the noises (N) and
interferences (FL) which includes RL that are present at the time of packet reception.
𝑆𝑁𝐼𝑅 = 𝑃𝑡𝑥− 𝑃𝐿𝑙𝑜𝑠𝑠∑𝑁+ ∑𝐹𝐿
(6.3)
151
Figure 6.20: Average Packet loss for 5-Vehicle Swarm against SNIR and Reverberation Levels at 50 m
Knowing this relationship between transmitter power, reverberation level and SNIR could be
beneficial in a vehicle for predicting optimal transmitter power levels in a similar way to the
feedback obtained in terrestrial networks. Optimising the energy levels used to send packets will
be beneficial for both reducing energy consumption as well as increasing success of its packets
being delivered without error. However, with increasing transmitter power, there is also the
potential to increase reverberation levels, which may then negate the benefit of increasing
transmitter power. This issue is a very interesting one and one that we believe will play an
important role in underwater short-range communications and will be investigated in future
work.
The average NCCP levels are provided in Figure 6.21 for the 5-vehicle network using 1 W and 3
W transmitter powers and the same set of sea state conditions of Figure 6.20. Here each dot
represents a different NCCP value based on a different Packet Arrival Rate. The dots with
higher NCCP in each vertical sequence towards the top of the graph have Packet Arrival Rates
that are operating well below cycle saturation as has been previously discussed. The smallest
NCCP value in each case is at cycle saturation and the cluster of results at that point is due to
the number of simulation that were done of packet arrival rates that are greater than the arrival
rate at cycle saturation. At higher SNIR levels the NCCP is equivalent to the ideal results
presented in Figure 6.8.
152
Figure 6.21: Average NCCP for 5-Vehicle Swarm at 50 m showing the increasing NCCP as Packet Arrival rate falls below cycle saturation and variations in Reverberation Levels
(Sea States)
As channel conditions worsen, the NCCP at cycle saturation (when it is at its lowest) increases
due to the regular missing of a packet in the NCCP exchange, which can be viewed similarly to
the missing of a packet in a cycle due to a lower packet arrival rate. This will force an increase
in the number of cycles required to complete an NCCP period when there are missing packets
which therefore means the average NCCP will increase at an increasing rate as shown.
For a 5-vehicle swarm operating at 50 m, these NCCP levels are well within the QoS boundary
(~3 s) requiring a NCCP of < 3 s. In Figure 6.21, even the networks operating with very low
packet arrival rates of 1 pkt/s are still within the QoS limits. This is not the case for a SNIR of 8.5
dB which was not included in Figure 6.21 as the NCCP at cycle saturation was at ~23 s and ~22
s for the 3 W and 1 W operations respectively. These NCCP values are significantly beyond the
QoS boundary and would not be able to exchange location information amongst each other
quickly enough for the AUV’s to operate as a swarm. The obvious solution again is to increase
the transmitter power, which in the case of the 1 W network, which is operating in medium sea
state, would improve the SNIR and come down the curve to a very low packet loss. However,
the question of the trade-off between increasing the vehicles transmitter power and the impact
that it would have on the reverberation levels that it is attempting to overcome remains to be
153
tested. Mechanisms to reduce the reverberation levels is the alternative approach and solutions
around antenna directionality might be useful.
6.6.3 Comparison with TDMA protocol and Channel Capacity Utilisation Benefits To examine a network working closer to the NCCP boundary, a 15-vehicle swarm operating at
50 m average range between sequenced vehicles, shown in Figure 6.6 (a), is presented in
Figure 6.22 under varying SNIR. A 3 W transmitter power is used. This section will also evaluate
the comparison between the AST-TDMA with Wait and the TDMA protocols in the non-ideal
underwater channel.
At high SNIR the networks are operating close to ideal with a minimum NCCP at cycle
saturation of 1.3 s and 2.66 s for the AST-TDMA and TDMA protocols respectively as shown in
Figure 5.11. Here the NCCP for packet arrival rates at and above cycle saturation are well
within the NCCP QoS boundary. However as packet arrival rate reduces the NCCP grows
above the NCCPlimit of 3 s.
With reducing SNIR, the minimum NCCP begins to increase as discussed in the 5-vehicle case
of Section 6.6.2. Here however the poorer channel conditions are responsible for pushing a
network beyond the QoS limitations and therefore would be too large to operate as a single
cluster.
Figure 6.22: Average NCCP at λsat for changes in SNIR due to Reverberation Levels: Comparison between AST-TDMA and TDMA protocols for 15-V swarm
0
1
2
3
4
5
6
12 13 14 15 16 17 18 19
NCC
P (s
)
SNIR (dB)
TDMA AST-TDMA
154
Figure 6.23: Channel Capacity Utilisation at λsat for changes in SNIR due to Reverberation Levels: Comparison between AST-TDMA and TDMA for a 15-V swarm
The TDMA protocol, as shown in ideal conditions, operates at higher NCCP compared to the
AST-TDMA protocol due to the large guard times required to avoid packet collisions. In this 15-
vehicle swarm, this means that there is little room to allow for poor channel conditions and at a
SNIR of 15 dB the minimum NCCP for the TDMA protocol goes over the 3 s limit. Thus, the
AST-TDMA protocol gives a much broader range of conditions to operate in for a 15-vehicle
swarm or conversely allows for a larger swarm to operate under the same conditions as was
shown in Figure 5.23.
The comparison of protocols based on Channel Capacity Utilisation illustrates the real benefits
of the adaptive protocols where the AST-TDMA protocol demonstrates its ability to use the
limited channel at twice the rate to the TDMA protocol even under poor channel conditions as
shown in Figure 6.23.
6.6.4 Variations in Packet Length Using a transmitter power of 3 W and a 5-vehicle swarm, the variations that will be experienced
in NCCP at λsat for changes in packet lengths are illustrated in Figure 6.28. The minimum NCCP
values for 424 bits can be seen in Figure 6.25 following the red results for each of the
reverberation levels. As packet length increases marginally there is a minor increase in the
NCCP for each of the sea state levels. For lower RL levels where SNIR is high and there are
minimal packet losses the NCCP for each packet length will only marginally increase due to the
additional transmission time required for the larger packet. The change that does occur between
packet sizes at these low RL is the packet arrival rate, which will decrease as packet size
increases as demonstrated in Figure 6.19. Thus, at lower sea states the opportunity to increase
packet size for exchanging additional data (in addition to the swarm formation requirements) will
0
10
20
30
40
50
60
12 13 14 15 16 17 18 19
Chan
nel C
apac
ity
Uti
lisai
ton
(%)
SNIR (dB)
AST-TDMA
TDMA
155
not have a large impact on NCCP and will in fact benefit from higher channel resource utilisation
as shown in Figure 6.20.
However, as the SNIR reduces as the reverberation levels increases, so with the packet losses
begin to increase. This forces NCCP to increase as vehicles may need to take several cycles of
data exchange before it obtains a complete set of data to input into its swarm formation
algorithm. It can be seen that with higher sea states, the increasing packet size has a greater
negative impact on NCCP and therefore the ability of the swarm to maintain swarm
synchronisation. The advantage of increasing packet size is suggested to enable piggybacking
of data so that each vehicle could support the delivery of other vehicles information and thus
potentially reduce the effects of packet losses. This raises several possible lines of investigate
that include; Does a vehicle need all other vehicles information or is it its closest neighbours
information that is enough. Also, if a piggybacking algorithm is to operate, which vehicles
information should it transmit with its own? Should it be the last information it received or should
it be based on knowledge of other vehicles need for data? Here the trade-off is related to
whether the increasing NCCP due to the additional cycles required due to the packet losses will
be able to be reduces by increasing packet size which increases NCCP in its own right.
6.6.5 Introduction of Swarm Reverberation In this section, the evaluation of the AST-TDMA MAC protocol will include the impact of
incorporating the proposed Swarm Reverberation, which was discussed in Chapter 3, Section
3.3.4.1 as a different form of reverberation to Sea-surface Reverberation that has been used as
the major interference factor for evaluation so far.
Figure 6.24: NCCP for variations in Data Packet Size for 5-V Swarm at 50 m
0
1
2
3
4
5
6
424 664 744 1224
NCC
P (s
) at C
ycle
Sat
urat
ion
Data Packet Size (bits)
RL: Low (-54 dB) RL: Medium (- 50 dB) RL: Medium High (-48 dB) RL: High (-46 dB)
156
Analysis of the impact that reverberation will have on the probability that a swarm vehicle can
successfully receive other vehicles communication based on transmitter power levels will be
presented. The expected Swarm Reverberation levels will be first compared to the Noise and
Sea-surface Reverberation Levels used in previous analysis. Then results on performance of
the AST-TDMA MAC protocol are presented with variations in transmitter power level, packet
length and density (range) of the swarm.
6.6.5.1 Noise and Reverberation Levels Using the channel parameters given in Table 6.2, a comparison between swarm reverberation
(where the reception of one and two reflections are included), sea-surface reverberation (based
on wind and sea state conditions) and ambient noise including background and self-noise
verses the average range between transmitting and receiving vehicles is presented in Figure
6.25. Swarm reverberation is demonstrated here to play a major role in the overall value of
interference (IF) experienced by a vehicle based on the potential overlap of reflections during
the reception of data packet. This value of reverberation is based on no loss of signal strength
on reflection and that the direct path echo will not be cancelled by other reflected waves
occurring in its path.
Swarm reverberation values are similar to the sea-surface reverberation fixed values used here.
The dominance of one form over another will depend on the many parameters with the
difference between range and depth being a significant one. Ambient noise, even at high wind
conditions, becomes much less significant compared with the reverberation levels proposed
here.
Figure 6.25: Noise and Reverberation Levels
20 25 30 35 40 45 50
-100
-90
-80
-70
-60
-50
-40
-30
Range (m)
Noi
se a
nd R
ever
bera
tion
Leve
ls (d
B)
Swarm reverberation, two reflections, Ptx=1WSwarm reverberation, one reflection, Ptx=1WSea-surface reverberation, High Sea StateSea-surface reverberation, Low Sea StateAmbient Noise (wind = 10 m/s)Ambient Noise (wind = 1 m/s)
157
Figure 6.26: Noise and Reverberation Levels for Variation in Transmitter Power
As expected, the closer the vehicles are operating the higher the swarm reverberation values
will be. In addition, the higher the transmitter power level the higher the swarm reverberation
level as illustrated in Figure 6.26.
Figure 6.26 also shows the combined values of noise and reverberation over increasing
average range between swarm vehicles and the difference in using a transmitted power level of
1 W and 3 W. The overall value of noise and interference is dominated by swarm reverberation
at lower ranges and at higher ranges for higher transmitter powers.
6.6.5.2 Variations in Transmitter Power The following OpNet results for packet loss incorporate swarm reverberation along with sea-
surface reverberation and ambient noise impacts on SNIR at increasing ranges.
Varying the transmission power levels are shown to impact more critically on packet loss than
packet size, as seen in Figure 6.27. As range increases, bigger packet sizes for the same
transmission power can also be seen to influence loss.
Figure 6.28 illustrates expected packet loss using the same packet size with variations in sea
state as well as transmitter power. It can be seen in Figure 6.25 that while we assume Sea-
Surface reverberation does not change with range compared to swarm reverberation which by
definition does, the dominance of sea-surface reverberation increase with range. This is
demonstrated particularly for high sea state in the ranges of interest. Thus, managing
transmitter power is an important element for MAC layer performance with relation to packet
loss with changes in reverberation levels.
20 25 30 35 40 45 50-50
-45
-40
-35
-30
Range (m)
Noi
se a
nd R
ever
bera
tion
Leve
ls (d
B)
Total Noise & Reverberation: Ptx = 3W, low sea & wind = 1 m/sSwarm reverberation, one reflection, Ptx = 3WTotal Noise & Reverberation: Ptx = 1W, high sea & wind = 10 m/sTotal Noise & Reverberation: Ptx = 1W, low sea & wind = 1 m/sSwarm reverberation, one reflection, Ptx = 1W
158
Figure 6.27: Packet Loss with variations in Packet Size and Transmitter Power
Figure 6.28: Packet Loss with variations in Transmitter Power and Sea State
6.7 Conclusion This chapter has continued the exploration into the operational and MAC layer parameters of
the new adaptive MAC protocol, AST-TDMA, proposed for closely working AUV’s in a swarm
arrangement and compared it with a conventional TDMA protocol in the same conditions. Non-
159
ideal channel conditions were introduced with the development of a OpNet pipeline model of the
acoustic channel for short-range communication.
It has been shown through simulations that the AST-TDMA protocols outperforms TDMA with its
ability to handle a larger number of vehicles within a network while maintaining QoS levels and
with significantly better channel resource utilisation. A new metric, Channel Capacity Utilisation
was introduced and along with the NCCP metric is able to evaluate the protocols ability to
maintain swarm synchronisation and channel resource utilisation.
Protocol modifications were proposed and investigated and the AST-TDMA with Wait
modification compared with the original AST-TDMA with Token, showed improved NCCP and
Channel Capacity Utilisation. Examination of various sizes of swarms and the impact that
changes in transmitter power and packet lengths will mean to the ability of a swarm to maintain
swarm synchronisation was presented. Variations to packet length were also examined to allow
additional payload data to be exchanged as well as the option to piggybacking of data as a
method to reduce the impact of higher packet losses.
The final section introduced the evaluation of Swarm Reverberation and the impact that various
forms of reverberation will have on the reception of a packet of data in an USSN. It was shown
that reverberation will have a significant impact on the reception of signals, due to the
interference level that reverberation can produce. Increasing the transmitter power was seen in
this research to enable more reliable communications in harsher underwater environments
using various densities of swarming vehicles, however there are trade-offs that need to be
considered, particularly in relation to the increase in reverberation levels that will occur when
swarm reverberation is taken into account.
Knowing the relationship between transmitter power, reverberation level and SNIR could be
beneficial in a vehicle for predicting optimal transmitter power levels in a similar way to the
feedback obtained in terrestrial networks. Optimising the energy levels used to send packets will
be beneficial for both reducing energy consumption as well as increasing success of its packets
being delivered without error. However, with increasing transmitter power, there is also the
potential to increase reverberation levels, which may then negate the benefit of increasing
transmitter power. This issue is a very interesting one and one that we believe will play an
important role in underwater short-range communications and will be investigated in future
work.
Along with the evaluation of the protocol performances and network operational limitations are
the many additional questions and problems that have been raised. There is still much work to
do before large groups of AUV’s will be operating in the oceans in a swarm-like fashion.
160
This page is intentionally blank
161
Chapter 7 Conclusion
7.1 Introduction The three key aspects of the research presented in this thesis are:
• Exploration of the fundamental parameters of a short-range underwater acoustic
communication channel;
• Classification and development of application architectures for underwater swarm
sensor networks which was presented as a new taxonomy; and
• Design of two new MAC layer protocols, for the two swarm deployment applications
established, that exploit the unique characteristics of the underwater acoustic
environment.
This thesis was organised into seven chapters and the main chapters and their contribution are
presented here as a summary of this research.
In Chapter 1 a brief description of the background of this work is presented together with the
main research objectives and key research contributions.
A taxonomy for Mobile Underwater Wireless Sensor Networks was developed and presented in
Chapter 2, which lead to the application classifications and deployment topologies of
Underwater Swarm Sensor Networks (USSN). The bio-inspired formation control algorithms and
swarm pattern formation control algorithms are presented which define the explicit data traffic
characteristics for the two deployment strategies and a discussion of the challenges of
underwater communication is presented. This lead to more specific research questions related
to the communication protocols that have been designed.
USSN require AUV’s (Autonomous Underwater Vehicles) to be working in close proximity to
each other, less than 50 m, which requires short-range communication that is non-conforming to
the typical applications that have been used in underwater environments. Thus the exploration
of the short-range underwater acoustic channel is presented in Chapter 3. Some
experimentation was conducted to analyse the short-range attenuation and compare this to the
attenuation models developed for the far field. Attenuation levels at short ranges were
measured but only for a very limited range due to the harsh operating environment of working
underwater. The important result here was the recognition of the potential reverberation levels
that may impact on short-range communication, where reverberation levels dominate over
ambient noise levels and therefore require a SNIR rather than a SNR.
In Chapter 4, an overview of the state of the art in underwater MAC layer protocols was
presented. In general, research and developments of underwater MAC layer protocols have
focused on adapting techniques from terrestrial networks but this does lead quickly to
diminishing returns due to the different operating conditions underwater. In addition, research
162
on underwater MAC layer protocols have focused on longer-range applications that have been
the common operational mode for ocean work. We instead needed to focus on the challenges
and opportunities posed by short-range applications. In this chapter the prospect of utilising the
spatial-temporal environment that is unique for underwater acoustic communications is
presented.
Two new time-based MAC protocols using tokens are introduced in Chapter 5 that utilise this
unique spatial-temporal condition underwater at short-range. The protocols incorporate the use
of ‘non exclusive channel access’ a method that is possible because of these unique channel
characteristics, which in particular are the low bandwidths and long propagation delays. These
adaptive protocols also have a significant advantage over other time-based protocols as they
avoid both the need for guard times and time synchronisation, which are both major drawbacks
in time based protocols for high latency environments. The Quality of Service (QoS) limitations
for swarm operations are developed and a new metric, NCCP (Neighbourhood Communication
Cycle Period), is proposed to test the new MAC protocol for the swarms’ ability to maintain
swarm synchronisation. The objective was to determine the maximum number of vehicles that
could be operated within a swarm based on different operating ranges and packet sizes.
In Chapter 6, an event-driven simulation environment was developed to test the new cluster
based swarm communication protocol in a non-ideal underwater channel. The objective here
was to investigate the impact that packet losses due to channel errors have on the adaptive
protocol and how transmitter power variations can influence the performance. Modifications to
the MAC protocol are discussed in relation to improvements to deal with packet losses due to
reception errors and due to information loss based on the inability of the network to exchange
packets fast enough. Variations in transmitter power and packet length are examined as a
means of overcoming these problems. Channel Capacity Utilisation is the second new metric
introduced to evaluate the ability of the protocol to maximise the channel resources. A
conventional TDMA protocol is used throughout the work as a comparison protocol.
It has been shown through simulations that the new adaptive MAC protocols outperform a
conventional TDMA protocol for swarm operations in their ability to disseminate information in a
time-critical manner and with higher channel capacity utilization. Using the new protocols also
allows a much higher density of vehicles to operate in a swarm like network.
7.2 Research Contributions As presented in Chapter 1, the key contributions from this work are summarised as the:
• Design of a SNIR algorithm that reflects the distinctive characteristics of a short-range
underwater acoustic communication channel and the reverberation interference that is
applicable when large numbers of AUV’s are operating in close proximity to each other
requiring long bursts of energy to transmit packets throughout the swarm;
• Implementation of this new channel model in OpNet simulation package for the
replication of a ‘realistic’ underwater acoustic operating environment;
163
• Development of a new Taxonomy for the classification of Mobile Underwater Wireless
Sensor Networks from which the Underwater Swarm Sensor Network (USSN) can be
defined. Based on this taxonomy two deployment topologies were defined with
associated data communication traffic requirements and QoS boundaries;
• Design and testing and analysis of two new MAC layer communication protocols for
USSN’s that utilise the unique spatial-temporal environment underwater and allow ‘non-
exclusive’ access to the channel while avoiding packet collisions; and
• Proposal of two new performance metrics for the analysis of the MAC layer protocols:
NCCP (Neighbourhood Communication Cycle Period) and Channel Capacity
Utilisation.
7.3 Future Research The following suggestions for future research are not exhaustive, but are focused on the major
areas of development that will need to be tackled before underwater autonomous swarms will
be a reality.
1. Understanding the reverberation level and dynamics in relation to the sending and receiving of long packets of data between vehicles operating at short-range. In this work, the recognition of reverberation as having a significant impact on the
reception of data packets due to the interference level that reverberation produces has
been proposed. This has not been fully tested and understood and would be a very
useful addition to the knowledge of SNIR levels in underwater short-range acoustic data
communications.
2. Relationship between transmitter power and reverberation levels. Increasing the
transmitter power was seen in this research to enable more reliable communications in
harsher underwater environments; however, the interference due to reverberation levels
which are in themselves self generated, are expected to increase with increasing
transmitter power. This relationship is not well understood and poses an interesting
problem about the trade-off between increased transmitter power levels and the impact
this has on increased reverberation levels. An understanding of this would support
better transmitter power management and reverberation levels in a communication
swarm network.
3. Clustering and use of smaller neighbourhoods. In this thesis, we consider a single
cluster swarm network, however for scalability purposes, the development of multi-
cluster swarms may be beneficial. The option of using CDMA between clusters was
proposed by Salva-Garau et. al. [113] and this concept appears to have some merit in
combination with the single cluster protocols proposed here. Alternatively the use of
transmitter power could be used to define clusters, as the attenuation in underwater
environments is large and could be sensitive enough to be used as a means of
segregation of clusters of swarm vehicles.
164
4. Algorithms for dynamic updating of vehicle sequencing. More specifically as an
improvement to the adaptive protocols proposed here is to develop an optimisation
algorithm around how the vehicles should be sequenced to maintain minimum NCCP.
This would also provide flexibility for vehicles to enter and exit a neighbourhood and to
dynamically scale the swarm size in response to changing applications and the external
environment.
5. Prioritisation of Data. The prioritisation of selected data (e.g. navigation data) for rapid
transmission around the swarm while allowing additional data to be transmitted if there
is channel resources available could be useful to support faster mission outcomes. This
prioritisation of data may be application dependent and dynamically adjusted as the
swarm environment changes. NCCP with piggybacking modification can also be further
investigated as a method to improve data transmission in the swarm. The use of larger
packet sizes will reduce the overheads for communication and this may be useful when
vehicles are operating in close proximity.
165
References
[1] Abramson, N. “The ALOHA System - Another Alternative for Computer
Communications.” Proc. 1970 Fall Joint Computer Conference. AFIPS Press. AFIPS
Press, 1970.
[2] Acar, G., and A.E Adam. “ACMENet: an underwater acoustic sensor network protocol
for real-time environmental monitoring in coastal areas.” IEE Proc. Radar, Sonar and
Navigation 153, no. 4 (August 2006).
[3] Ahn, Joon, Affan Syed, Bhaskar Krishnamachari, and John Heidemann. “Design and
analysis of a propagation delay tolerant ALOHA protocol for underwater networks.” Ad
Hoc Networks Journal (Elsevier ), 2010.
[4] Akyildiz, I, D Pompili, and T Melodia. “State of the Art in Protocol Research for
Underwater Acoustic Sensor Networks.” ACM Mobile Computing and Communciation
Review, Oct 2007.
[5] Akyildiz, Ian, and Mehmet Vuran. Wireless Sensor Networks. John Wiley & Sons, 2010.
[6] Arkin, Ronald C. “Cooperation without communication: Multiagent schema-based robot
navigation.” Jounral of Robotic Systems (John Wiley & Sons) 9, no. 3 (1992): 351-364.
[7] Arkin, Ronald C. “Motor Schema - Based mobile robot navigation.” The International
Journal of Robotics Research (Sage Journals) 8, no. 4 (August 1989): 92-112.
[8] Bean, T., J. Canning, G. Beidler, M. O'Rourke, and D. Edwards. “Language and logic to
enable collaborative behaviour among multiple autonomous underwater vehicles.”
International Journal of Intelligent Control and Systems 13 (2008): 67-80.
[9] Benson, B., et al. “Design of a low cost, underwater acoustic Modem for short-range
sensor networks.” IEEE OCEANS. Sydney: IEEE, 2010. 1-9.
[10] Berni, Alessandro, Diego Merani, John Potter, and Robert Ben. “Heterogeneous
System Framework for Underwater Networking.” MILCOM2011 - Military
Communication Conference 2011. 2011. 2050-2056.
[11] Bertsekas, Dimitri, and Robert Gallager. Data Networks. Edited by Margaret Rizzi.
Prentice Hall, 1992.
[12] Biaz, Saad, and Yiming Ji. “A survey and comparison on localisation algorithms for
wireless ad hoc networks.” International Journal of Mobile Communication
(Inderscience Enterprises Ltd.) 3, no. 4 (December 2005): 374-410.
[13] Brancart, Claude. “AUV Challenges for 2020.” IEEE Oceanic Engineering Society
Newsletter, 2005.
[14] Browning, D, J Everett, R Mellen, and W Thorp. “Attenuation of Low-Frequency Sound
in Freshwater.” Science 6 162, no. 3858 (1968): 1120-1121.
[15] Burrowes, Gunilla, and Jamil Y. Khan. “Adaptive Token Polling MAC Protocol for
Wireless Underwater Networks.” International Symposium on Wireless & Pervasive
Computing. Melbourne, 2009.
166
[16] Burrowes, Gunilla, and Jamil Y. Khan. “Short-range Underwater Acoustic
Communication Networks.” In Autonomous Underwater Vehicles, by Nuno A Cruz, 173-
198. Croatia: InTech, 2011.
[17] Burrowes, G., Brown J., Khan, J. "Adaptive Space Time - Time Division Multiple Access
Protocol (AST - TDMA) for an Underwater Swarm of AUV's". IEEE OCEANS, Bergen,
2013
[18] Cao, Y. Uny, Alex S Fukunaga, and Andrew B Kahng. “Cooperative Mobile Robotics:
Antecedents and Directions.” Autonomous Robots (Kluwer Academic Publishers) 4
(1997): 1-23.
[19] Carbonelli, Cecilia, and Urbashi Mitra. “Cooperative Multihhop Communication for
Underwater Acoustic Networks.” WUWNet'06. LA, USA: ACM, 2006. 97-100.
[20] Caruthers, J. Fundamentals of Marine Acoustics. Elsevier Scientific Publishing, 1977.
[21] Chan, Wah-Chun. Performance Analysis of Telecommunications and Local Area
Networks. Kluwer Academic Publishers, 2000.
[22] Chen, Baozhi, Dario Pompili, and Manish Parashar. “Bio-inspired communications for
coordination among autonomous underwater vehicles.” IEEE Sarnoff Symposium.
IEEE, 2010. 1-6.
[23] Chen, W, and U Mitra. “Packet scheduling for multihopped underwater acoustic
communication networks.” IEEE OCEANS'07. IEEE Publishers, 2007.
[24] Chirdchoo, N., Wee-Seng Soh, and Kee Chua. “Aloha-based MAC Protocls with
Collision Avoidance for Underwater Acoustic Networks.” Preceedings for IEEE
INFOCOM 2007. IEEE, 2007. 2271-2275.
[25] Chirdchoo, N., Wee-Seng Soh, and Kee Chua. “RIPT: A Receiver-Initiated Reservation-
Based Protocol for Underwater Acoustic Networks.” IEEE Journal on Selected Areas in
Communications (IEEE JOURNALS & MAGAZINES) 26, no. 9 (2008): 1744-1753.
[26] Chitre, Mandar, Shiraz Shahabudeen, Lee Freitag, and M Stojanovic. “Recent
advances in aderwater acoutic communications and networking.” IEEE OCEANS. IEEE,
2008.
[27] Coates, Rodney F.W. Underwater Acoustic Systems. New York: Halsted Press, div of
John Wiley & Sons, 1989.
[28] CoCoRo. CoCoRo COllective Cognitive Robots. Thomas Schmickl. 2011. CoCoRo
COllective Cognitive Robots (accessed May 2013).
[29] Cox, A.W. Sonar and Underwater Sound. Lexington Books, 1974.
[30] Crowther, Will, R. Rettberg, D. Walden, S. Ornstein, and E. Heart. “A System for
Broadcast Communication: Reservation-ALOHA.” Proceedings of the 6th Hawaii
International Conference on Systems Sciences. Honolulu, 1973. 371-374.
[31] Cuff, T.R, and R.W. Wall. “Support platform and communications to manage
cooperative AUV operations.” IEEE OCEANS. Asia Pacific: IEEE, 2006. 1-8.
167
[32] Cui, Jun-Hong, Jiejun Kong, Mario Gerle, and Shengli Zhou. “The challenges of building
scalable mobile underwater wireless sensor networks for aquatic applications.” IEEE
Network, May/June 2006: 12-18.
[33] Daladier, J.M., and M.A. Labrador. “A data link layer in support of swarming of
autonomous underwater vehicles.” IEEE OCEANS. Europe, 2009.
[34] De, Swades, Priyatosh Mandal, and Shyam Chakraborty. “On the characterization of
Aloha in underwater wireless networks.” Journal of Mathematical and Computer
Modelling, 2011: 2093-2107.
[35] Deneubourg, J, S Aron, S Goss, and J Pasteel. “The self-organizating exploratory
pattern of the argentine ant.” Journal of Insect Behavior (Plenum Publishing
Corporation) 3, no. 2 (1990).
[36] Diamant, Roee, and Lutz Lampe. “Spatial reuse Time-Division Multiple Access for
broadcast Ad Hoc underwater acoustic communication networks.” IEEE Journal of
Oceanic Engineering (IEEE) 36, no. 2 (April 2011): 172-185.
[37] Domingo, M. “Overview of channel models for underwater wireless communication
networks.” Physical Communications, 2008: 163-182.
[38] Dorigo, M, and L Gambardella. “Ant colony system: a cooperative learning approach to
the travelling salesman problem.” IEEE Transaction on Evolutionary Comuputation
(IEEE Publishing) 1, no. 1 (1997): 53-66.
[39] Doukkali, Hayat., Loutfi Nuaymi, and Sebastien Houcke. “Power and distance based
MAC algorithm for underwater acoustic networks.” IEEE OCEANS. 2006.
[40] Dressler, Falko, and Ozgur Akan. “Bio-inspired networking: From Theory to Practice.”
IEEE Communication Magazine, November 2010.
[41] Dunbabin, M, J Roberts, K Usher, G Winstanley, and P Corke. “A hybrid auv design for
shallow water reef navigation.” International Conference on Robotics and Automation
(ICRA). 2005. 2117-2122.
[42] Eckart, C. Principles of Underwater Sound. National Research Council, University of
California, 1952
[43] Egbo, H.N., N.L. Hallin, P.L. Ray, M. O'Rourke, J.F. Frenzel, and D.B. Edwards.
“Token-Based Medium Access Control Solution For Underwater Acoustic Broadcast.”
IEEE OCEANS 2009. IEEE CONFERENCE PUBLICATIONS, 2009.
[44] Etter, P. Underwater Acoustic Modelling and Simulation. Spoon Press, 2003.
[45] Favaro, F., P. Casari, F. Guerra, and M. Zorzi. “Data upload from a static Underwater
network to an AUV: Polling or random access?” IEEE OCEANS. Yeosu: IEEE, 2012. 1-
6.
[46] FENWICK, J.W., P.M. NEWMAN, and J.J. LEONARD. “Cooperative concurrent
mapping and localization .” IEEE Int. Conf. on Robotics and Automation . Washington
DC: IEEE, 2002. 1810-1817.
[47] Fisher, F, and V Simmons. “Sound Absorption in sea water.” Journal of the Acoustical
Society of America 62, no. 3 (1977)
168
[48] Franck, L. Performance comparison of mobile ad-hoc and delay tolerant networks.
2011. http://www-public.int-evry.fr/.
[49] Franco, E., L. Magni, T, Parisini, M.M. Polycarpou, and D.M. Raimondo. “Cooperative
constrained control of distributed agents with nonlinear dynamices and delayed
information exchange: A stabilising receding-horizon approach.” IEEE Transaction
Automous Control (IEEE) 53, no. 1 (February 2008): 324-338.
[50] Francois, R, and G Garrison. “Sound absorption based on ocean measurements: Part 1
and 2.” Journal of the Acoustical Society of America 72, no. 3 & 6 (1982): 896-907,
1879-1890.
[51] Gheitanchi, Shahin, Falah Ali, and Elias Stipidis. “Particle swarm optimisaiton for
adaptive resource allocation in communication networks.” EURASIP Journal on
Wireless Communications and Networking (Hindawi Publishing Corporation) 2010
(2010): 13.
[52] Guerra, F., P. Casari, A. Berni, J. Potter, and M. Zorzi. “Performance evaluation of
random and handshake-based channel access in collaborative moobile underwater
networks.” IEEE OCEANS. 2010.
[53] Guo, Xiaoxing, M.R. Frater, and M.J. Ryan. “Design of a propagation-delay tolerant
MAC protocol for underwater acoustic sensor networks.” IEEE Journal of Oceanic
Engineering 34, no. 2 (2009): 170-180.
[54] Guo, Zheng, G. Colombo, Bing Wang, Jun-Hong Cui, D. Maggiorini, and G.P. Rossi.
“Adaptive Routing in Underwater Delay/Disruption Tolerant Sensor Networks.” Fifth
Annual Conference on Wireless on Demand Network Systems and Services, 2008.
WONS 2008. . IEEE, 2008. 31-39.
[55] Hajenko, T, and C Benson. “The high frequency underwater acoustic channel.” IEEE
OCEANS 2010. Sydney: IEEE Publishers, 2010. 1-3.
[56] Heidemann, John, Milica Stojanovic, and Michele Zorzi. “Underwater sensor networks:
applicatons, advances and challanges.” Philosophical Transactions of the Royal Society
A (Royal Society Publishing), no. 370 (November 2011): 158-175.
[57] Heidemann, John, Wei Ye, Jake Wills, Affan Syed, and Yuan Li. “Research Challenges
and Applicaitons for Underwater Sensor Networking.” IEEE Wireless Communication
and Networking Conference. Las Vegas, US, 2006.
[58] Herman, Ted, and Sebastien Tixeuil. “A distributed TDMA slot assignment algorithm for
wireless sensor networks.” In ALGOSENSORS, by S Nikoletseas and J. Rolim, 45-58.
Berlin Heidelberg: Springer-Verlag, 2004.
[59] Hinchey, Michael, Roy Sterritt, and Chris Rouff. “Swarms and Swarm Intelligence.”
Computer (IEEE Publishing), April 2007: 111-113.
[60] Holmes, J, W Carey, J Lynch, A Newhall, and A Kukulya. “An autonomous underwater
vehicle towed array for ocean acoustic measurements and inversions.” IEEE
OCEANS'05. Europe: IEEE Publishers, 2005. 1058-1061.
169
[61] Hsu, Chih-Cheng, Kuang-Fu Lai, Cheng-Fu Chou, and Kate Ching-Ju Lin. “ST-MAC:
spatial-temporal MAC scheduling for underwater sensor networks.” IEEE INFOCOM.
2009.
[62] Huang, Pai-Han, Ying Chen, Bhaskar Krishnamachari, and Anil Kumar. “Link
scheduling in a single broadcast domain underwater networks.” IEEE International
Conference on Sensor Networks, Ubiquitous and Trustworthy Computing. 2010.
[63] IEEE, Webmaster [email protected]. IEEE 802.15 WPAN Task Group 4 (TG4). IEEE. 27
January 2010. http://www.ieee802.org/15/pub/TG4.html.
[64] International, Hydro. “Acoustic Modem Product Survey.” Hydro International, June
2007: 36-39.
[65] Jabba, Daladier, and Miguel Labrador. “A data link layer in support of swarming of
autonomous underwater vehicles.” IEEE OCEANS 2009 - EUROPE. IEEE
CONFERENCE PUBLICATIONS, 2009. 1-10.
[66] Janik, V. "Acoustic communication networks in marine mammals" Animal
Communication Networks. 2005 McGregor, Cambridge University Press, p. 390-415
[67] Joerg, Kalwa, et al. “The MORPH Concept and Its Application in Marine Research .”
IEEE OCEANS . Bergen Norway: MTS/IEEE, 2013. 1-8.
[68] Johnson, M, J Preisig, L Freitag, and M Stojanovic. “FSK and PSK performance of the
utility acoustic modem.” IEEE OCEANS'99. IEEE Publishers, 1999. 1512.
[69] Johnson, R. University Corporation of Atmospheric Research. National Earth Science
Teachers Association. 2012.
www.windows.ucar.edu/tour/link/earth/Water/overview.html (accessed May 2013).
[70] Jonson, Thomas, Jonah Pezeshki, Victor Chao, Kristofer Smith, and James Fazio.
“Application of delay tolerant networking (DTN) in Airborne Networks.” MILCOM 2008 -
Military Communications Conference. IEEE, 2008. 1-7.
[71] Joordens, M.A. “Consensus Control for a System of Underwater Swarm Robots.” IEEE
Systems Journal (IEEE) 4, no. 1 (March 2010): 65-73.
[72] Kalantar, Shahab, and Uwe R. Zimmer. “Optima Localization by Vehicle Formations
Imitating the Nelder-Mead Simplex Algorithm.” Autonomous Robots , 2009.
[73] Karl, H., and Willig, A. Protocols and Architectures for Wireless Sensor Networks. John
Wiley & Sons, 2007.
[74] Katz, Ian. “A Delay-Tolerant Networking Framework for Mobile Underwater Acoustic
Network.” International Symposium on Unmanned Untethered Submersible Technology.
2007.
[75] Kennedy, J, and R Eberhart. “Particule swarm optimization.” Proceedings of IEEE
International Conference on Neural Networks IV. IEEE, 1995. 1942-1948.
[76] —. Swarm Intelligence. Morgan Kaufmann, 2001.
[77] Kinsey, J., Eustice, R., and Whitcomb, L. “A Survey of underwayer vehicle navigation:
recent advantages and new challenges". Proceedings of the IFAC Conference of
Manoeuvering and Control of Marine Craft. IFAC , 2006.
170
[78] Kinsler, L, A Frey, A Coppens, and J Sanders. Fundementals of Acoustics. John Wiley
and Sons, 1982.
[79] Kleunen, Wonter van, Nirvana Meratnia, and Paul Havinga. “A set of simplified
scheduling constraints for underwater acoustic MAC scheduling.” Workshops of
International Conference on Advanced Information Networking and Applications. 2011.
[80] Kredo, K., P. Djukic, and P. Mohapatra. “STUMP: Exploiting position diversity in the
staggered TDMA underwater MAC protocol.” IEEE Infocom. 2009. 2961 - 2965 .
[81] Kredo, Kurtis, and Prasant Mohapatra. “Distributed scheduling and routing in
underwater wireless networks.” Global Telecommunications Conference (GLOBECOM
2010),. 2010.
[82] Kulkarni, Indraneel, and Dario Pompili. “Task allocation for networked autonomous
underwater vehicles in critical missions.” IEEE Journal on Selected Areas in
Communications (IEEE) 28, no. 5 (June 2010): 716-727.
[83] Leon-Garcia, Alberto, and Indra Widjaja. Communication Networks. Fundamental
Concepts and Key Architectures. Thomas Casson, 2000.
[84] LinkQuest. SoundLink Underwater Acoustic Modems. July 2012. www.link-quest.com
(accessed May 2013).
[85] Maffei, A.R, K Fall, and D.N Chayes. “Ocean Instrument Internet: Using Disruption
Tolerant Networking to Join Heterogeneous Oceanographic Instrumentation into a
Single Network.” Ocean Science Meeting. 2006.
[86] Mare, J. “Path following algorithm for minimally specified lawn-mower type AUV
missions.” IEEE OCEANS 2010 . Sydney : IEEE CONFERENCE PUBLICATIONS,
2010. 1-5.
[87] Maurelli, F., Z. Saigol, J. Cartwright, D.M. Lane, A. Bourque, and B. Nguyen. “TDMA-
based exchange policies for multi-robot communication of world information .”
Proceedings of IFAC MCMC 2012 - Manoeuvring and Control of Marine Craft . 2012.
[88] Melodia, T., H. Kulhandjian, L. Kuo, and E. Demirors. “Advances in Underwater
Acoustic Networking.” In Mobile Ad Hoc Networking: Cutting Edge Directions, by M.
Conti, S. Giordano and I. Stojmenovic S. Basagni, 804-852. John Wiley and Sons, Inc.,
Hoboken, NJ, 2013.
[89] Molins, M., and M. Stojanovic. “Slotted FAMA: a MAC protocol for underwater acoustic
networks.” IEEE OCEANS. Asia-Pacific, 2006.
[90] MONSUNII. MONSUN II - Simulated energy load balancing in a v-formation.
www.youtube.com/watch?v=gT9I3xRAQn4 (accessed May 2013).
[91] Nasri, N, A Kachouri, L Andrieux, and M Samet. “Design considerations for wireless
underwater communication transceiver.” International Conference on Signal, Circuits
and Systems. Monastir, 2008.
[92] Nathaniel Bowditch, LL.D. The American Practical Navigator, An Epitome of Navigation.
National Geospatial-Intelligence Agency, Bethesda, Maryland: National Imagery and
Mapping Agency, 2002.
171
[93] Navarro, I, and F Matia. “An introduction to swarm robotics.” ISRN Robotics, 2013.
[94] Noh, Youngtae, Paul Wang, Uichin Lee, Dustin Torres, and Mario Gerla. “DOTS: A
propagation delay-aware opportunistic MAC protocol for underwater sensor networks.”,
18th IEEE International Conference on Network Protocols (ICNP) . 2010.
[95] Osterioh, Christoph, and Erik Maehle. “Low-power microcontroller-based acoustic
modem for underwater robot communication.” 41st International Symposium on
Robotics and Conference on ROBOTIK 2010. IEEE, 2010.
[96] Osterioh, Christoph, Pionteck Thilo, and Erik Maehle. “MONSUN II: A small and
inexpensive AUV for underwater swarms.” Proceedings of ROBORIK 2012. IEEE, 2012.
[97] Ottman, W.A.F.W, B.P Amavasai, S.P McKibbin, and F Caparrelli. “An analysis of
collective movement models for robotic swarms.” EUROCON 2007 The International
Conference on Computer as a Tool. Warsaw, 2007. 2373-2380.
[98] Paley, Derek A, Fumin Zhang, and Leonard Ehrich. “Cooperative control for ocean
sampling: THe glider coordinated control system.” IEEE Transactions on Control
Systems Technology (IEEE) 16, no. 4 (July 2008): 735-744.
[99] Park, Min Kyoun, and V. Rodoplu. “UWAN-MAC: An energy-efficient MAC protocol for
underwater acoustic wireless sensor networks.” Ocean Engineering 32, no. 3 (2007):
710-720.
[100] Parrish, N, S Roy, W Fox, and P Arabshahi. “Rate-range for an FH-FSK Acoustic
Modem.” WUWNet'07. Montreal: ACM, 2007. 93-96.
[101] Partan, Jim, Jim Kurose, and Brian Neil Levine. “A survey of practical issues in
underwater networks.” WUWNet'06. Los Angeles: ACM, 2006.
[102] Peleato, B., and M. Stojanovic. “Distance aware collision avoidance protocol for ad-hoc
underwater acoustic sensor networks.” IEEE Communications Letters, 2007: 1025-
1027.
[103] Petillo, Stephanie, Henrik Schmidt, and Arjuna Balasuriya. “Constructing a distributed
AUV network for underwater plume-tracking operations.” International Journal of
Distributed Sensor Networks (Hindawi Publishing Corporation) 2012 (2012): 12.
[104] Petrioli, C., R. Petroccia, and M. Stojanovic. “A comparative performance evaluation of
MAC protocols for underwater sensor networks.” IEEE OCEANS. 2008.
[105] Pompili, Dario. Efficient Communication Protocols for Underwater Acoustic Sensor
Networks. Thesis, School of Electrical and Computer Engineering, Georgia Institute of
Technology, ProQuest Information and LEarning Company, 2007, 191.
[106] Pompili, Dario, and Ian Akyildiz. “Overview of networking protocol for underwater
wireless communications.” IEEE Communication Magazine, January 2009: 97-102.
[107] Preisig, J., "Acoustic Propagation Considerations for Underwater Acoustic
Communication Network Development", WUWNet Los Angeles 2006
[108] Qu, Z., J. Wang, R.A. Hull, and J. Martin. “Cooperative control design and stability
analysis for multi-agent systems with communication delays.” IEEE International
Conference on Robotics and Automation. IEEE, 2006. 970-975.
172
[109] Rappaport, T. Wireless Communications, Principles and Practices. New Jersey:
Prentice Hall PTR, 1996.
[110] Reynolds, Craig. “Flocks, Herds, and Schools: A Distributed Behavioral Model.”
SIGGRAPH '87 Conference Proceedings (Computer Graphics) (ACM) 21, no. 4 (1987):
25-34.
[111] Roberts, Lawrence. “ALOHA Packet System With and Without Slots and Capture.”
Computer Communications Review, April 1975: 28-42.
[112] Sahin, Erol. “Swarm Robotics: From Sources of Inspiration to Domains of Application.”
Swarm Robotics. Vol. 3342. Edited by Erol Sahin and William Spears. Springer, 2005.
10-20.
[113] Salva-Garau, F., and M. Stojanovic. “Multi-cluster protocol for ad hoc mobile underwater
acoustic networks.” Proceedings OCEANS 2003. IEEE CONFERENCE
PUBLICATIONS, 2003. 91-98.
[114] Schill, Felix, and Uwe R. Zimmer. “Distributed Dynamical Omnicast Routing.” Journal of
Complex Systems (Complex Systems Publications, Inc) 16 (2006): 299-316.
[115] Schill, Felix, and Uwe Zimmer. “Pruning Local Schedules for Efficient Swarm
Communication.” Underwater Technology Symposium UT07. Tokyo: IEEE Publishing,
2007. 594-600.
[116] Schmickl, T, et al. “CoCoRo - The Self-aware Underwater Swarm.” 5th IEEE
COnference on Self-Adaptive and Self-Organizing Systems Workshop. IEEE
Publishing, 2011. 120-126.
[117] Seah, W.K.G., Hwee-Xian Tan, Zheng Liu, and V.M.H. Ang. “Multiple-UUV approach for
enhancing connectivity in underwater ad-hoc sensor networks.” IEEE OCEANS. 2005.
2263-2268.
[118] Sehgal, A, I Tumar, and J Schonwalder. “Variability of available capacity due to the
effects of depth and temperature in the underwater acoustic communication channel.”
IEEE OCEANS. Europe: IEEE Publishers, 2009. 1-6.
[119] Shahabudeen, S., M. Chitre, and M. Motani. “A multi-channel MAC protocl for AUV
networks.” IEEE OCEANS. Europe, 2007.
[120] Smith, S.M., J.C. Park, and A. Neel. “A peer-to-peer communication protocol for
underwater acoustic communication.” IEEE OCEANS. 1997. 268-272.
[121] Sozer, Ethem M., Milica Stojanovic, and John G. Proakis. “Underwater Acoustic
Networks.” IEEE Journal of Oceanic Engineering (IEEE) 25, no. 1 (January 2000).
[122] Speller, Luke. “Fishing for Pollutants (SHOAL Project)” Engineering World,
Febuary/March 2013.
[123] Stojanovic, M., L. Freitag, J. Leonard, and P. Newman. “A network protocol for multiple
AUV localisation.” IEEE OCEANS. 2002. 604-611.
[124] Stojanovic, Milca. “Underwater acoustic communications.” In Wiley Encyclopedia of
Electrical and Electronic Engineering. John Wiley & Sons, 2003.
173
[125] Stojanovic, M. “On the Relationship between Capacity and Distance in an Underwater
Acoustic Communication Channel.” WUWNet'06. ACM, 2006.
[126] Stojanovic, Milca. “Uderwater Acoustic Communication Tutorial.” IEEE OCEANS.
Bergen: IEEE, 2013.
[127] Sullivan, E, and M Taroudakis. Handbood of Signal Processing in Acoustics. Vol. 2.
RSpringer, 2008.
[128] Syed, a, W Ye, B Krishnamachari, and J Heidemann. “Understanding spatio-temporal
uncertainty in medium access with ALOHA protocols.” Preceedings of the Second ACM
International Workshop on Underwater Networks (WUWNet07). Quebec: ACM, 2007.
41-48.
[129] Syed, A.A., Wei Ye, and J. Heidemann. “T-Lohi: A new class of MAC protocols for
underwater acoustic sensor networks.” INFOCOM 2008. The 27th Conference on
Computer Communications. IEEE . 231-235.
[130] Syed, A.A., Ye Wei, and J. Heidemann. “Comparison and Evaluation of the T-Lohi MAC
for Underwater Acoustic Sensor Networks.” IEEE Journal on Selected Areas in
Communications (IEEE Journals and Magazines) 26, no. 9 (2008): 1731-1743.
[131] Syed, Affan, and John Heidemann. “Time synchronisation for high latency acoustic
networks.” IEEE INFOCOM. Barcelona, 2006.
[132] Thorp, W. H. “Deep-ocean sound attenuation in the sub- and low-kilocycle-per-second
region.” Journal of the Acoustical Society of America (Acoustical Society) 38, no. 4
(1965): 648-654.
[133] Tourrilhes, Jean. The MAC level (link layer). Sponsored by Hewlett Packard. 2004.
http://www.hpl.hp.com/personal/Jean_Tourrilhes/Linux/Linux.Wireless.mac.html
(accessed 2012).
[134] Urick, Robert J. Principles of Underwater Sound for Engineers. McGraw-Hill Book
Company, 1967.
[135] Vasilescu, I., Kotay K., Rus D., Dunbabin M., and Corke P. 2005. "Data collection,
storage, and retrieval with an underwater sensor network." Proceedings of the 3rd
International Conference On Embedded Networked Sensor Systems, ACM SenSys, pp.
154–165, New York, NY, USA
[136] Vicsek, Tamas, A Czirok, Eshel Ben-Jacob, Inon Chen, and Ofer Shochet. “Noval type
of phase transition in a system of self driven particles.” Physical Review Letters, 7
August 1995: 1226-1229.
[137] Waite, A.D. Sonar for Practicing Engineers. West Sussex: John Wiley & Sons, 2005.
[138] Walree, P. “Acoustic Modems: Product survey.” Hydro International Magazine, June
2007: 36-39.
[139] Warthman, Forrest. “Introducing-delay-disruption-tolerant-networking-dtn.”
InterPlanetary Networking Special Interest Group (IPNSIG). 2013.
http://ipnsig.org/about-us/ (accessed 2013).
174
[140] WeiXing, Feng, Wang KeJun, Ye XiuFen, and Guo ShuXiang. “Novel algorithms for
coordination of underwater swarm robotics.” International Conference on Mechatronics
and Automation. Luoyang, China: IEEE, 2006. 654-659.
[141] Wilensky, Uri. NetLogo flocking model. Northwestern University. 1998.
http://ccl.northwestern.edu/netlogo/ (accessed May 2013).
[142] Wills, J., Ye, W., Heidemann, J., "Low-Power Acoustic Modem for Dense Underwater
Sensor Networks", WUNet 2006
[143] Xie, Peng, and Jun-Hong Cui. “R-MAC: An energy-efficient MAC protocol for
underwater sensor networks.” Wireless Algorithms, Systems and Applications, 2007.
WASA 2007. International Conference on. 2007. 187 - 198 .
[144] Yackoski, J, and C Shen. “UW-FLASHER: Achieving high channel utilisation in a time-
based acoustic MAC protocol.” Preceedings of the Thrid ACM International Workshop
on Underwater Networks (WUWNet08). San Francisco: ACM, 2008.
[145] Yang, Yongming, and Yantao Tian. “Swarm robots aggregation formation control
inspired by fish school.” IEEE International Conference on Robotics and Biomeimetics.
Sanya, China: IEEE, 2007. 805-809.
[146] Yoon, Seokhoon, and Chunming Qiao. “Cooperative Search and Survey Using
Autonomous Underwater Vehicles (AUVs).” IEEE Transactions on Parallel and
Distributed Systems (IEEE Journals & Magazines) 22, no. 3 (2011): 364-379.
[147] Zhao, Yi Zhi, Chunyan Miao, and Maode Ma. “Performance of Adaptive Scheduling
MAC (AS-MAC) Protocol with Different AS-Period in Multi-hop Networks.” IEEE 6th
Conference on Industrial Electronics and Applications (IEEE) (2011): 1881-1886.
[148] Zhang, Wenyi, Milica Stojanovic, and Urbashi Mitra. “Analysis of linear multihop
underwater acoustic network.” IEEE OCEANS Engineering (IEEE) 35, no. 4 (October
2010): 961-970.
[149] Zimmer, Uwe, R. Serafina. Directore RISISE. 2 November 2006.
http://serafina.anu.edu.au (accessed May 2013).
175
Appendix A Fisher & Simmons [47] Coefficients
𝛼 =𝐴1𝑓1𝑓3
𝑓12 + 𝑓2+𝐴2𝑃2𝑓2𝑓3
𝑓22 + 𝑓2+ 𝐴3𝑃3𝑓2𝑚−1
𝐴1 = (1.03 × 10−3 + 2.36 × 10−10𝑇 − 5.22 × 10−12𝑇2)𝑠𝑒𝑐 𝑚−1
𝑓1 = 1.32 × 103(𝑇 + 273.1) exp �−1700
𝑇 + 273.1�𝐻𝑧
𝐴2 = (5.62 × 10−6𝑇)𝑠𝑒𝑐 𝑚−1
𝑓2 = 1.55 × 107(𝑇 + 273.1) exp �−3052
𝑇 + 273.1�𝐻𝑧
𝑃2 = 1 − 10.3 × 10−4𝑃 + 3.7 × 10−7𝑝2
𝐴3 = (55.9 − 2.37𝑇 + 4.77 × 10−2𝑇2 − 3.48 × 10−4𝑇3)𝑥10−15 𝑠𝑒𝑐2𝑚−1
𝑃3 = 1 − 3.84 × 10−4𝑃 + 7.57 × 10−8𝑃2
where f is in Hz, T is in degrees centigrade and P is in atm.
176
Appendix B MatLab Code
Underwater Acoustic Channel: SINR Calculations using Thorps Absorption Coefficients
function SINR = SINR(r) % Function for calculating the SINRdB % Author: Gunilla Elizabeth Burrowes January 2014 % Ptxelec (Watts) % f is freq (kHz) & % r range (m) % w is wind speed (m/s) 1m/s = 2knots % s is shipping factor (0 to 1) % k is the spreading factor k=2 spherical and k=1 cylindrical % B bandwidth (Hz) % R data rate(bps) % n efficiency of reciever (%) Ptxelec = 1; f = 40; r = 20:1:50; B = 5000; R = 9600; k = 2; w = 1; s = 0; n = 0.3; NswarmdB = 10; %dB NreverbdB = -45; %dB for i = 1:length(r) AC = ((0.11*f^2)/(1+f^2) + (44*f^2)/(4100+f^2)+ 0.000275*f^2 + 0.0033); %Thorp TL(i) = 10*k*log10(r(i)) + AC*r(i)*10^(-3); %Path Loss SLTxdB(i) = 170.8 + 10*log10(Ptxelec) + 10*log10(n); SLRxadB(i) = SLTxdB (i) + 10*log10(n) - TL(i); %Rxpower in dB re1uPa Prxaco (i) = SLRxadB (i) - 170.8 ; %receiver power in dB Prxelec(i) = (10^(Prxaco (i)/10)); %receiver power in W Nturb = 17 - 30*log10(f); Nship = 40 + 20*(s-0.5) + 26*log10(f) - 60*log10(f+0.03); Nwind = 50 + 7.5*w^0.5 + 20*log10(f) - 40*log10(f+0.4); Nthermal = -15 + 20*log10(f); Nswarm = 10^((NswarmdB-170.8)/10); bkg_noise = 6.3 * 10^(-18);
177
NOISEtotal = 10^(( Nturb - 170.8 )/10) + 10^(( Nship - 170.8)/10) + 10^((Nwind - 170.8)/10) + 10^((Nthermal - 170.8)/10); NOISEtotalB = (NOISEtotal + Nswarm + bkg_noise)*B; %total noise in watts accross bandwidth Nreverb = 10^((NreverbdB)/10); accumnoise = NOISEtotalB + Nreverb; SINR (i) = Prxelec(i)/ accumnoise; SINRdB (i) = 10*log10(SINR(i)); end plot (r,SINRdB)
178
Appendix C bpsk Modulation Curve
179
Appendix D Energy Consumption in an Autonomous Underwater Vehicle
180
Appendix E Process Model OpNet Code
/* Process model C form file: AST_TDMA_5nodes_N1_wait.pr.c */ /* Portions of this file copyright 1986-2011 by OPNET Technologies, Inc. */ =========================== NOTE ========================== This file is automatically generated from AST_TDMA_5nodes_N1_wait.pr.m during a process model compilation. Do NOT manually edit this file. Manual edits will be lost during the next compilation. =========================== NOTE ========================== #include <string.h> /* OPNET system definitions */ #include <opnet.h> /* Header Block */ #define UPPER_IN_STRM_INDEX 1 #define LOWER_IN_STRM_INDEX 0 #define LOWER_OUT_STRM_INDEX 0 #define UPPER_STRM ( \ (op_intrpt_type() == OPC_INTRPT_STRM) && \ (op_intrpt_strm() == UPPER_IN_STRM_INDEX)) #define LOWER_STRM ( \ (op_intrpt_type() == OPC_INTRPT_STRM) && \ (op_intrpt_strm() == LOWER_IN_STRM_INDEX)) #define SLOTSTART (op_intrpt_type() == OPC_INTRPT_SELF) #define TX_TIME_WAIT (op_intrpt_type() == OPC_INTRPT_SELF) #define PKT_TIME_WAIT (op_intrpt_type() == OPC_INTRPT_SELF) #define subq_index 0 #define QEMPTY (op_subq_stat(0, OPC_QSTAT_PKSIZE) == 0.0) /* Global Variables */ int subm_pkts; int subm_pkts_rx; int subm_tokenpkts; int subm_tokenpkts_rx; double timeofcycle; int total_slots; /* End of Header Block */ /* State variable definitions */ /* State Variables */ Stathandle throughput ; Objid node_id ; Stathandle pktsreceived ; int rcvd_pkts ; Stathandle offeredload ; Packet * tx_message ; Packet * rcvd_message ; Packet * qpacket ; Packet * invalid_tx_message ;
181
int invalid_tx_pkts ; Stathandle invalidtxpkts ; Packet * dpacket ; int d_pkts ; Stathandle destroyed_pkts_tx ; int q_pkts ; Stathandle queuedpkts ; Stathandle ete_pkt ; double ete_delay ; int Tx_number ; Stathandle totalsentpkts ; OpT_Packet_Size pkt_len ; double pkt_drate ; double prop_delay ; int no_collisions ; int pkt_from_tx_one ; int pkt_from_tx_two ; int pkt_from_tx_three ; int pkt_from_tx_four ; int pkt_from_tx_five ; int pkt_from_tx_six ; int pkt_from_tx_seven ; int pkt_from_tx_eight ; Stathandle pkts_tx_one ; Stathandle pkts_tx_two ; Stathandle pkts_tx_three ; Stathandle pkts_tx_four ; Stathandle pkts_tx_five ; Stathandle pkts_tx_six ; Stathandle pkts_tx_seven ; Stathandle pkts_tx_eight ; int sent_pkts ; int Sender_number ; Packet * cpacket ; int c_pkts ; Stathandle collided_pkts_rx ; Stathandle pkts_sent_node ; double cur_time ; double rx_time ; int num_nodes ; int max_packet_count ; Stathandle NCCP_Time ; double NCCPstart ; double NCCPstop ; double NCCPeriod ; int Pkt_Rcvd[20] ; int pkt_queue ; Stathandle slot_utilise ; Stathandle channel_utilise ; double range ; Stathandle Range ; Stathandle pkts_tx_nine ; Stathandle pkts_tx_ten ; int pkt_from_tx_nine ; int pkt_from_tx_ten ; double EndofRx ; double slot_time ; double av_slot_time ; Stathandle Slot_Times ; double starttime ;
182
double time_sent ; double cycle_time ; double max_prop_delay ; Packet * token ; int sent_tokenpkts ; double NCCPdelay ; double NCCPcreat ; Stathandle pktsize ; int collisionsRx ; Stathandle NCCP_creation ; Stathandle node_channel_utilise ; int pkt_size ; Stathandle PDelay ; int pktsizetoken ; Stathandle pkt_size_token ; int rcvd_tokenpkts ; Stathandle tokens_rxed ; Stathandle tokens_sent_node ; Stathandle Ave_slottime ; Stathandle Interfer_pkt ; int node_slots ; Stathandle slot_utilise_token ; int rxpkt_len ; int NCCPcounter ; Stathandle NCCPave ; Stathandle No_slots ; Stathandle No_NCCP ; Stathandle node_success_rate ; Stathandle NumberVehicles ; Stathandle data_gen_rate ; Stathandle beta_ratio ; Stathandle Interarrival ; double rangeave ; Stathandle Zeta ; double NCCPdifference ; Stathandle NCCP_difference ; Stathandle pkt_service ; Stathandle density ; Stathandle ave_range ; double StartofRx ; double StartofTx ; double EndofTx ; double rangeseq ; Stathandle Range_Seq ; Stathandle Pkt_Age ; double Pkt_age_ave ; double creat_time1 ; double creat_time2 ; double creat_time3 ; double creat_time4 ; double creat_time5 ; double creat_time6 ; double creat_time7 ; double creat_time8 ; double creat_time9 ; double creat_time10 ; double creat_time11 ; double service_delay ; Stathandle pkt_per_cycle ; Stathandle Ave_CycleTime ;
183
Stathandle pkt_in_cycle ; double pktincycle ; double slot_length_seq ; Stathandle Slot_Lengthseq ; Stathandle No_cycles ; Stathandle A_Cycle_Time ; int total_discards ; Stathandle Total_Discards ; int generated_pkts ; int generated_pkts_cycle ; Stathandle gen_pkt_cycle ; Stathandle gen_pkt ; Stathandle cycles_per_NCCP ; char data_rate ; int NCCP ; double creat_time12 ; double creat_time13 ; double creat_time14 ; double creat_time15 ; int pkt_from_tx_eleven ; int pkt_from_tx_twelve ; int pkt_from_tx_thirteen ; int pkt_from_tx_fourteen ; int pkt_from_tx_fifteen ; double waittimestart ; double waittimestop ; Stathandle waittime ; int DestroyedPkts ; Stathandle Destroyed_Pkts ; int TotalDestroyedPkts ; /* Function Block */ /*Transmitter Side ...*/ static void SendPacket (void) { FIN (SendPacket()); /* Get packet from queue after flushing it, so most recent packet sent*/ while (op_subq_stat(0, OPC_QSTAT_PKSIZE) > 1) { invalid_tx_message = op_subq_pk_remove (0, OPC_QPOS_HEAD); ++invalid_tx_pkts; op_pk_destroy (invalid_tx_message); } generated_pkts_cycle = 1 + invalid_tx_pkts; generated_pkts = generated_pkts + generated_pkts_cycle; total_discards = total_discards + invalid_tx_pkts; op_stat_write (invalidtxpkts, (double) invalid_tx_pkts); op_stat_write (gen_pkt_cycle, (double) generated_pkts_cycle); invalid_tx_pkts = 0; generated_pkts_cycle = 0; tx_message = op_subq_pk_remove (0, OPC_QPOS_HEAD); op_pk_fd_set(tx_message, 0, OPC_FIELD_TYPE_INTEGER, Tx_number, 0);
184
op_pk_send (tx_message, LOWER_OUT_STRM_INDEX); pkt_size = op_pk_total_size_get(tx_message); /*Packets sent from all Nodes*/ ++subm_pkts; /*Packets sent from this Node*/ ++sent_pkts; /*Slots from all Nodes*/ ++total_slots; /*Slot from this Node*/ ++node_slots; /*set time packet sent*/ time_sent = op_sim_time(); StartofTx = op_sim_time(); FOUT } static void NCCPcalculation (void) { FIN (NCCPcalculation()); if (NCCP == 1) { /*Initialize Booleans' for each node*/ Pkt_Rcvd[0] = 1; Pkt_Rcvd[1] = 0; Pkt_Rcvd[2] = 0; Pkt_Rcvd[3] = 0; Pkt_Rcvd[4] = 0; NCCP = 0; NCCPstart = op_td_get_dbl (tx_message, OPC_TDA_RA_START_TX); NCCPcreat = op_pk_creation_time_get (tx_message); //StartofTx; //or } else { } FOUT } static void sendtoken (void) { FIN (sendtoken()); /* Create token packet*/ token = op_pk_create(104); op_pk_fd_set(token, 0, OPC_FIELD_TYPE_INTEGER, Tx_number, 0); op_pk_send (token, LOWER_OUT_STRM_INDEX); pktsizetoken = op_pk_total_size_get(token); pkt_size = 104; /*Packets sent from all Nodes*/ ++subm_tokenpkts; /*Packets sent from this Node*/ ++sent_tokenpkts;
185
/*Slots from all Nodes*/ ++total_slots; /*Slots from this Node*/ ++node_slots; /*set time packet sent*/ time_sent = op_sim_time(); FOUT } static void QFrame (void) { FIN (QFrame()); qpacket = op_pk_get(UPPER_IN_STRM_INDEX); op_subq_pk_insert (0, qpacket, OPC_QPOS_TAIL); ++q_pkts; pkt_queue = op_q_stat (OPC_QSTAT_IN_PKSIZE); FOUT; } static void destroypkt (void) { FIN (destroypkt()); dpacket = op_pk_get(LOWER_IN_STRM_INDEX); op_pk_destroy(dpacket); ++d_pkts; FOUT; } /*Receiver Side*/ /* This function gets the received packet, destroys */ /* it, and logs the incremented received packet total */ static void rcvd_pkt (void) { FIN (rcvd_pkt()); /* Get packet from bus receiver input stream */ rcvd_message = op_pk_get (LOWER_IN_STRM_INDEX); op_pk_fd_get(rcvd_message, 0, &Sender_number); pkt_drate = op_td_get_dbl(rcvd_message, OPC_TDA_RA_TX_DRATE); rxpkt_len = op_pk_total_size_get(rcvd_message); prop_delay = op_td_get_dbl(rcvd_message, OPC_TDA_RA_END_PROPDEL); range = op_td_get_dbl(rcvd_message, OPC_TDA_RA_END_DIST); ete_delay = (op_td_get_dbl(rcvd_message, OPC_TDA_RA_END_RX)) + 2*((double)pkt_len)/100000 - op_pk_creation_time_get(rcvd_message); service_delay = (op_td_get_dbl(rcvd_message, OPC_TDA_RA_END_RX)) + 2*((double)pkt_len)/100000 - (op_td_get_dbl(rcvd_message, OPC_TDA_RA_START_TX)); rangeave = rangeave + range; StartofRx = op_td_get_dbl(rcvd_message, OPC_TDA_RA_START_RX); EndofRx = op_td_get_dbl(rcvd_message, OPC_TDA_RA_END_RX) + 2*((double)pkt_len)/100000; op_stat_write (PDelay, (double) prop_delay); op_stat_write (Range, (double) range);
186
op_stat_write (ete_pkt, (double) ete_delay); op_stat_write (pkt_service, (double) service_delay); ++node_slots; /* Increment the count of received packet */ ++rcvd_pkts; ++subm_pkts_rx; pktincycle = pktincycle + 1; if (Sender_number == 2) { ++pkt_from_tx_two; if (pkt_from_tx_two > 1) {creat_time2 = 0;} Pkt_Rcvd[1] = 1; creat_time2 = op_pk_creation_time_get(rcvd_message); } else if (Sender_number == 3) { ++pkt_from_tx_three; if (pkt_from_tx_three > 1) {creat_time3 = 0; } Pkt_Rcvd[2] = 1; creat_time3 = op_pk_creation_time_get(rcvd_message); } else if (Sender_number == 4) { ++pkt_from_tx_four; if (pkt_from_tx_four >1) {creat_time4 = 0;} Pkt_Rcvd[3] = 1; creat_time4 = op_pk_creation_time_get(rcvd_message); } else if (Sender_number == 5) { ++pkt_from_tx_five; if (pkt_from_tx_five >1) {creat_time5 = 0; } Pkt_Rcvd[4] = 1; creat_time5 = op_pk_creation_time_get(rcvd_message); } if ((NCCP==0) && ((Pkt_Rcvd[1] == 1) && (Pkt_Rcvd[2] == 1) && (Pkt_Rcvd[3] == 1) && (Pkt_Rcvd[4] == 1))) { NCCPstop = EndofRx; NCCPeriod = NCCPstop - NCCPstart; NCCPdelay = NCCPstop - NCCPcreat; NCCPdifference = NCCPdelay - NCCPeriod; NCCP = 1; ++NCCPcounter;
187
Pkt_age_ave = ((NCCPstop - creat_time2) + (NCCPstop - creat_time3) + (NCCPstop - creat_time4) + (NCCPstop - creat_time5) )/(double)(num_nodes-1); DestroyedPkts = (pkt_from_tx_two + pkt_from_tx_three + pkt_from_tx_four + pkt_from_tx_five ) - (num_nodes - 1); op_stat_write (Destroyed_Pkts, (double) DestroyedPkts); op_stat_write (Pkt_Age, (double) Pkt_age_ave); op_stat_write (NCCP_Time, (double) NCCPeriod); op_stat_write (NCCP_creation, (double) NCCPdelay); op_stat_write (NCCP_difference, (double) NCCPdifference); op_stat_write (pkts_tx_one, (double) pkt_from_tx_one); op_stat_write (pkts_tx_two, (double) pkt_from_tx_two); op_stat_write (pkts_tx_three, (double) pkt_from_tx_three); op_stat_write (pkts_tx_four, (double) pkt_from_tx_four); op_stat_write (pkts_tx_five, (double) pkt_from_tx_five); TotalDestroyedPkts = DestroyedPkts + TotalDestroyedPkts; DestroyedPkts = 0; Pkt_age_ave = 0; pkt_from_tx_one = 0; pkt_from_tx_two = 0; pkt_from_tx_three = 0; pkt_from_tx_four = 0; pkt_from_tx_five = 0; } FOUT; } /* This function writes the end-of-simulation channel */ /* traffic and channel throughput statistics to a */ /* vector file */ static void record_stats (void) { FIN (record_stats()); cur_time = op_sim_time(); /* Record final statistics */ op_stat_write (Slot_Times, (double) slot_time); op_stat_write (ave_range, (double) rangeave/(rcvd_pkts)); op_stat_write (No_slots, (double) total_slots); op_stat_write (NumberVehicles, (double) num_nodes); op_stat_write (No_cycles, (double) total_slots/num_nodes); op_stat_write (cycles_per_NCCP,(double) (total_slots/num_nodes)/NCCPcounter); op_stat_write (Ave_CycleTime, (double) ((cur_time-10)/total_slots)*num_nodes); op_stat_write (Ave_slottime, (double) (cur_time-10)/total_slots); op_stat_write (pkt_per_cycle, (double) (rcvd_pkts)/(total_slots/num_nodes)); op_stat_write (No_NCCP, (double) NCCPcounter); op_stat_write (NCCPave,(double) (cur_time-10)/NCCPcounter); op_stat_write (density, (double) 10); op_stat_write (data_gen_rate, (double) q_pkts/(double)(cur_time - 10));
188
op_stat_write (Interarrival, (double) sent_pkts/(cur_time - 10)); op_stat_write (queuedpkts, (double) q_pkts); op_stat_write (destroyed_pkts_tx, (double) d_pkts); op_stat_write (totalsentpkts, (double) (subm_pkts - sent_pkts)); op_stat_write (pktsize, (double) pkt_len); op_stat_write (pkts_sent_node, (double) sent_pkts); op_stat_write (pktsreceived,(double) rcvd_pkts); op_stat_write (offeredload,(double) ((((subm_pkts - sent_pkts) * pkt_len) / (cur_time - 10)) /(double) pkt_drate)); op_stat_write (throughput,(double) ((((rcvd_pkts - (((rcvd_pkts/(double)NCCPcounter) - (num_nodes - 1))*NCCPcounter)) *pkt_len) / (double)(cur_time - 10)) / (double)pkt_drate)); op_stat_write (channel_utilise, (double) (((rcvd_pkts*(pkt_len/(double)pkt_drate)) + (sent_pkts*(pkt_len/(double)pkt_drate)))/(double)(cur_time - 10))*100); op_stat_write (node_channel_utilise, (double) ( ( ((NCCPcounter*(num_nodes - 1))*(pkt_len/(double)pkt_drate)) + (sent_pkts*(pkt_len/(double)pkt_drate)))/(double)(cur_time - 10))*100); op_stat_write (slot_utilise, (double) ((rcvd_pkts + sent_pkts)/ total_slots)*100); op_stat_write (node_success_rate, (double) ((rcvd_pkts + sent_pkts)/ total_slots)*100); op_stat_write (beta_ratio, (double) (200*pkt_drate)/(1500*pkt_len)); op_stat_write (Total_Discards, (double) total_discards); op_stat_write (gen_pkt, (double) generated_pkts); FOUT; } /* End of Function Block */ /** state (init) enter executives **/ { /* Get the maximum packet count, */ /* set at simulation run-time */ op_ima_sim_attr_get_int32 ("max packet count", max_packet_count); node_id = op_topo_parent(op_id_self()); op_ima_obj_attr_get (node_id, "user id", &Tx_number); /* Get the number of nodes. */ num_nodes = (op_topo_object_count (OPC_OBJTYPE_NODE_FIX)-2); /* Get the Propagation time, */ /* set at simulation run-time */ op_ima_sim_attr_get_dbl ("prop delay (s)", &max_prop_delay); NCCPeriod = 0; pkt_len = 424; //1224; //744; //664; //424; pkt_size = 424; //1224; //744; //664; //424; pkt_drate = 9600; /*initalising variables*/ /* schedule first interrupt for this process */ /* based on Transmission Time + prop delay + processing time */
189
starttime = 10.0; op_intrpt_schedule_self (starttime, 0); /* Initialize accumulator*/ NCCP = 1; Pkt_Rcvd[0] = 0; Pkt_Rcvd[1] = 0; Pkt_Rcvd[2] = 0; Pkt_Rcvd[3] = 0; Pkt_Rcvd[4] = 0; rcvd_tokenpkts = 0; subm_tokenpkts_rx = 0; subm_pkts = 0; subm_tokenpkts = 0; subm_pkts_rx = 0; rcvd_pkts = 0; invalid_tx_pkts = 0; d_pkts = 0; q_pkts = 0; c_pkts = 0; sent_pkts = 0; sent_tokenpkts = 0; node_slots = 0; total_slots = 0; NCCPcounter = 0; rangeave = 0; pktincycle = 0; slot_time = 0; timeofcycle = 0; total_discards = 0; generated_pkts_cycle = 0; generated_pkts = 0; slot_length_seq = max_prop_delay; DestroyedPkts = 0; TotalDestroyedPkts = 0; time_sent = 11; pkt_from_tx_one = 0; pkt_from_tx_two = 0; pkt_from_tx_three = 0; pkt_from_tx_four = 0; pkt_from_tx_five = 0; Range_Seq = op_stat_reg ("Sequence Range (m)", OPC_STAT_INDEX_NONE, OPC_STAT_GLOBAL); Slot_Lengthseq = op_stat_reg("Slot Length Sequence (s)", OPC_STAT_INDEX_NONE, OPC_STAT_GLOBAL); PDelay = op_stat_reg ("Progation Delay", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); Slot_Times = op_stat_reg("Total Cycle Time (s)", OPC_STAT_INDEX_NONE, OPC_STAT_GLOBAL); Range = op_stat_reg ("Range (m)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); ave_range = op_stat_reg ("Average range to other vehicles (m)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); No_slots = op_stat_reg("Total Number of Slots", OPC_STAT_INDEX_NONE, OPC_STAT_GLOBAL); No_cycles = op_stat_reg("Total Number of Cycles", OPC_STAT_INDEX_NONE, OPC_STAT_GLOBAL);
190
Ave_CycleTime = op_stat_reg("Average Cycle Time", OPC_STAT_INDEX_NONE, OPC_STAT_GLOBAL); Ave_slottime = op_stat_reg("Average Length of Slots (s)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); A_Cycle_Time = op_stat_reg("One Cycle Time", OPC_STAT_INDEX_NONE, OPC_STAT_GLOBAL); pkt_per_cycle = op_stat_reg("Packet per Cycle", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); pkt_in_cycle = op_stat_reg("Packets in Each Cycle", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); ete_pkt = op_stat_reg ("Pkt ETE Delay", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); pkt_service = op_stat_reg ("Packet Service Time", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); Destroyed_Pkts = op_stat_reg ("Destroyed Packets", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); NCCP_Time = op_stat_reg("NCCP Time", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); NCCP_creation = op_stat_reg("NCCP with Delay from Creation", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); NCCPave = op_stat_reg("Average NCCP Time", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); No_NCCP = op_stat_reg("Number of NCCP cycles", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); data_gen_rate = op_stat_reg("Sensor Data Generation Rate (pkts/s)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); Interarrival = op_stat_reg("Average Interarrival rate", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); NumberVehicles = op_stat_reg ("Number of Vehicles", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); density = op_stat_reg ("Swarm Density m2/vehicle", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); pktsize = op_stat_reg("Packet Size (bits)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); pkt_size_token = op_stat_reg("Packet Size Token (bits)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); totalsentpkts = op_stat_reg("Packets Sent from other Nodes (pkts)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); //tokens_rxed = op_stat_reg ("Tokens Rcvd at Node", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); //tokens_sent_node = op_stat_reg ("Tokens Sent from Node", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); pkts_sent_node = op_stat_reg ("Packets Sent from Node", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); pktsreceived = op_stat_reg("Successful Packets Rx (pkts)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); Total_Discards = op_stat_reg("Total Queue Discards (pkts)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); invalidtxpkts = op_stat_reg("Queue Discards (pkts)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); destroyed_pkts_tx = op_stat_reg("Number of pkts destroyed in MAC", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); queuedpkts = op_stat_reg("Queued Pkts (pkts/sec)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); collided_pkts_rx = op_stat_reg("Number of pkts collide while Rx", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); Interfer_pkt = op_stat_reg ("Interfer Pkts Successful", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); NCCP_difference = op_stat_reg("NCCP Difference (delay - min)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL);
191
Pkt_Age = op_stat_reg ("Average Packet Age", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); pkts_tx_one = op_stat_reg("Packets From Tx One", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); pkts_tx_two = op_stat_reg("Packets From Tx two", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); pkts_tx_three = op_stat_reg("Packets From Tx three", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); pkts_tx_four = op_stat_reg("Packets From Tx four", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); pkts_tx_five = op_stat_reg("Packets From Tx five", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); slot_utilise = op_stat_reg("Slot Utilisation - Successful Pkts Only", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); //slot_utilise_token = op_stat_reg("Slot Utilisation includes tokens", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); channel_utilise = op_stat_reg("Channel Utilisation", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); node_channel_utilise = op_stat_reg("Node Channel Utilisation", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); node_success_rate = op_stat_reg("Success Rate using node slots (%)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); offeredload = op_stat_reg("Offered Load G (normalised)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); throughput = op_stat_reg("Throughput S (normalised)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); Zeta = op_stat_reg("Zeta Ratio", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); beta_ratio = op_stat_reg("Beta Ratio", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); gen_pkt_cycle = op_stat_reg("Gen Pkt per Cycle (pkt)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); gen_pkt = op_stat_reg("Gen Pkt (pkt)", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); cycles_per_NCCP = op_stat_reg("Cycles per NCCP", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); waittime = op_stat_reg("Wait times", OPC_STAT_INDEX_NONE, OPC_STAT_LOCAL); } /** state (recieving) exit executives **/ FSM_STATE_EXIT_FORCED (6, "receiving", "AST_TDMA_5nodes_N1_wait [receiving exit execs]") FSM_PROFILE_SECTION_IN ("AST_TDMA_5nodes_N1_wait [receiving exit execs]", state6_exit_exec) { if ((Sender_number + 1) == 6) { rangeseq = op_td_get_dbl(rcvd_message, OPC_TDA_RA_END_DIST); op_stat_write (Range_Seq, (double) rangeseq); op_stat_write (pkt_in_cycle, (double) pktincycle); pktincycle = 1; slot_length_seq = EndofRx - op_td_get_dbl(rcvd_message, OPC_TDA_RA_START_TX); op_stat_write (Slot_Lengthseq, (double) slot_length_seq);
192
slot_time = slot_time + slot_length_seq; timeofcycle = timeofcycle + slot_length_seq; op_stat_write (A_Cycle_Time, (double) timeofcycle); timeofcycle = 0; op_intrpt_schedule_self ((EndofRx), 0); } /*Destroy the received packet */ op_pk_destroy (rcvd_message); }